Daily Papers

Automated vulnerability detection research has made substantial progress, yet its real-world impact remains limited. Prior work found that current vulnerability datasets suffer from issues including label inaccuracy rates of 20%-71%, extensive duplication, and poor coverage of critical Common Weakness Enumeration (CWE). These issues create a significant generalization gap where models achieve misleading In-Distribution (ID) accuracies (testing on splits from the same dataset) by exploiting spurious correlations rather than learning true vulnerability patterns. To address these limitations, we present a three-part solution. First, we introduce BenchVul, which is a manually curated and balanced test dataset covering the MITRE Top 25 Most Dangerous CWEs, to enable fair model evaluation. Second, we construct a high-quality training dataset, TitanVul, comprising 38,548 functions by aggregating seven public sources and applying deduplication and validation using a novel multi-agent LLM pipeline. Third, we propose a Realistic Vulnerability Generation (RVG) pipeline, which synthesizes context-aware vulnerability examples for underrepresented but critical CWE types through simulated development workflows. Our evaluation reveals that In-Distribution (ID) performance does not reliably predict Out-of-Distribution (OOD) performance on BenchVul. For example, a model trained on BigVul achieves the highest 0.703 ID accuracy but fails on BenchVul's real-world samples (0.493 OOD accuracy). Conversely, a model trained on our TitanVul achieves the highest OOD performance on both the real-world (0.881) and synthesized (0.785) portions of BenchVul, improving upon the next-best performing dataset by 5.3% and 11.8% respectively, despite a modest ID score (0.590). Augmenting TitanVul with our RVG further boosts this leading OOD performance, improving accuracy on real-world data by 5.8% (to 0.932).

Cross-modal drone navigation remains a challenging task in robotics, requiring efficient retrieval of relevant images from large-scale databases based on natural language descriptions. The RoboSense 2025 Track 4 challenge addresses this challenge, focusing on robust, natural language-guided cross-view image retrieval across multiple platforms (drones, satellites, and ground cameras). Current baseline methods, while effective for initial retrieval, often struggle to achieve fine-grained semantic matching between text queries and visual content, especially in complex aerial scenes. To address this challenge, we propose a two-stage retrieval refinement method: Caption-Guided Retrieval System (CGRS) that enhances the baseline coarse ranking through intelligent reranking. Our method first leverages a baseline model to obtain an initial coarse ranking of the top 20 most relevant images for each query. We then use Vision-Language-Model (VLM) to generate detailed captions for these candidate images, capturing rich semantic descriptions of their visual content. These generated captions are then used in a multimodal similarity computation framework to perform fine-grained reranking of the original text query, effectively building a semantic bridge between the visual content and natural language descriptions. Our approach significantly improves upon the baseline, achieving a consistent 5\% improvement across all key metrics (Recall@1, Recall@5, and Recall@10). Our approach win TOP-2 in the challenge, demonstrating the practical value of our semantic refinement strategy in real-world robotic navigation scenarios.

Computational knowledge graphs assign philosophical concepts to traditions based on corpus frequency: the school that mentions a concept most becomes its attributed tradition. We argue this conflates three measurements: textual power, historical priority, and philosophical significance, demonstrated using the darshana-graph, a knowledge graph of 28,322 relationships across Hindu, Buddhist, and Jain traditions. Seven of the top 25 concepts by betweenness centrality predate their attributed school by 288 to 2,288 years. Moksha, attributed to Advaita Vedanta, appears first in Jain sources over 1,200 years earlier. The most reliable snapshot, at 300 BCE using only explicitly dated sources, shows a genuinely pluralistic structure: 59% Vedic, 24% Jain, 18% Buddhist. We also quantify a critical distortion in the temporal method: between 300 CE and 800 CE the network grows from 18 to 1,028 nodes, with 97.4% carrying Advaita proxy dates, revealing that apparent dominance reflects textual survival, not philosophical history. Beyond correcting attribution bias, the temporally grounded graph enables structural homology analysis across traditions. Ego-network feature vectors applied to 48 temporally labelled concepts across eight traditions identify cross-tradition concept pairs with high structural similarity. The method recovers known correspondences including purusha-jiva (Samkhya/Jain, sim 0.990) and prakriti-maya (Samkhya/Vedic, sim 0.972), and surfaces novel homologies. Nibbana and samsara score 0.954 despite being doctrinal opposites: both function as the ultimate reference concept in their tradition's soteriology. Cetana (Buddhist intention) and ajiva (Jain non-living matter) score 0.923, a pairing absent from the literature. These are not claims of doctrinal equivalence but of measurable structural homology: different philosophical vocabularies navigating a shared conceptual space.

AI-associated lexical shifts have been documented mainly in Scientific English. We extend this work to 34 languages in the WMT News Crawl corpus, refining a split-halves continuation diagnostic that compares GPT-4.1 continuations with matched human gold-standard text. For each language, we derive ranked AI-overused lemmas using log prevalence ratios. We find substantial cross-lingual semantic convergence: semantically related concepts recur across typologically diverse languages, with 'emphasize'-type verbs appearing in 24 of 34 languages. Embedding-based and manual analyses support this pattern. We also examine diachronic uptake in news writing before and after ChatGPT's release. Tracking each language's top 20 AI-overused items, we find prevalence increases in 26 of 34 languages from 2020-2021 to 2023-2024, with a mean change of +15.1%, whilst matched baseline words show no comparable increase (-4.5%). In 10 languages with longer historical coverage, longitudinal analyses show post-2022 increases that exceed the modest shifts observed in earlier periods, though with smaller effect sizes than in Scientific English. We validate our approach extensively, including across seeds, model variants, data sizes, model families, and more. Our findings are consistent with the view that AI-associated lexical preferences extend beyond English and may exert cross-lingual homogenising pressure on global language use.

This article presents an experiment in fine-tuning a pretrained causal language model (Meta's Llama 3.1 8B Instruct) for aiding in three fundamental tasks of philological research: chronological and geographic attribution as well as text restoration in ancient Greek inscriptions and documentary papyri. Using a prompt-based instruct approach, the fine-tuned models surpass the state of the art in key metrics. For inscriptions, the models achieve a lower average character error rate (CER) of 22.5% (vs. 26.3%), while closely matching top-1 accuracy (60.9% vs. 61.8%) and top-20 accuracy (77.5% vs. 78.3%) for sequences up to 10 characters. They also provide a practical advantage by ignoring spaces during reconstruction, aligning better with the scriptio continua typically used in ancient written artifacts. In geographic attribution, the model outperforms previous benchmarks with a top-1 accuracy of 75.0% (vs. 70.8%) and a top-3 accuracy of 83.7% (vs. 82.1%). For dating, it achieves an average deviation of 26.2 years (vs. 29.3) and a median deviation of 1 year (vs. 3) from the actual date range. The models also set new baselines for documentary papyri, with a CER of 16.3%, a top-1 accuracy of 71.3%, and top-20 of 85.0% in text reconstruction; a top-1 accuracy of 66.4% and top-3 of 79.9% in geographic attribution; and, in chronological attribution, a deviation of 21.7 years from the actual termini post/ante quem, with a median deviation of 0 years.

There is burgeoning interest in designing AI-based systems to assist humans in designing computing systems, including tools that automatically generate computer code. The most notable of these comes in the form of the first self-described `AI pair programmer', GitHub Copilot, a language model trained over open-source GitHub code. However, code often contains bugs - and so, given the vast quantity of unvetted code that Copilot has processed, it is certain that the language model will have learned from exploitable, buggy code. This raises concerns on the security of Copilot's code contributions. In this work, we systematically investigate the prevalence and conditions that can cause GitHub Copilot to recommend insecure code. To perform this analysis we prompt Copilot to generate code in scenarios relevant to high-risk CWEs (e.g. those from MITRE's "Top 25" list). We explore Copilot's performance on three distinct code generation axes -- examining how it performs given diversity of weaknesses, diversity of prompts, and diversity of domains. In total, we produce 89 different scenarios for Copilot to complete, producing 1,689 programs. Of these, we found approximately 40% to be vulnerable.

Expert parallelism has been introduced as a strategy to distribute the computational workload of sparsely-gated mixture-of-experts (MoE) models across multiple computing devices, facilitating the execution of these increasingly large-scale models. However, the All-to-All communication intrinsic to expert parallelism constitutes a significant overhead, diminishing the MoE models' efficiency. Current optimization approaches offer some relief, yet they are constrained by the sequential interdependence of communication and computation operations. To address this limitation, we present a novel shortcut-connected MoE architecture with overlapping parallel strategy, designated as ScMoE, which effectively decouples communication from its conventional sequence, allowing for a substantial overlap of 70% to 100% with computation. When compared with the prevalent top-2 MoE architecture, ScMoE demonstrates training speed improvements of 30% and 11%, and inference improvements of 40% and 15%, in our PCIe and NVLink hardware environments, respectively, where communication constitutes 60% and 15% of the total MoE time consumption. On the other hand, extensive experiments and theoretical analyses indicate that ScMoE not only achieves comparable but in some instances surpasses the model quality of existing approaches in vision and language tasks.

Peripheral blood smears remain a cornerstone in the diagnosis of hematological neoplasms, offering rapid and valuable insights that inform subsequent diagnostic steps. However, since neoplastic transformations typically arise in the bone marrow, they may not manifest as detectable aberrations in peripheral blood, presenting a diagnostic challenge. In this paper, we introduce cAItomorph, an explainable transformer-based AI model, trained to classify hematological malignancies based on peripheral blood cytomorphology. Our data comprises peripheral blood single-cell images from 6115 patients with diagnoses confirmed by cytomorphology, cytogenetics, molecular genetics, and immunophenotyping from bone marrow samples, and 495 healthy controls, eight coarse classes. cAItomorph leverages the DinoBloom hematology foundation model and aggregates image encodings via a transformer-based architecture into a single vector. It achieves an overall accuracy of 0.72 in eight disease classification, with F1 scores of 0.76 for acute leukemia, 0.80 for myeloproliferative neoplasms and 0.94 for healthy cases. The overall accuracy increases to 0.87 in top-2 predictions. cAItomorph achieves high sensitivity for acute leukemia cases in external test sets. By analyzing attention heads, we demonstrate clinically relevant cell-level attentions in both internal and external test sets. Moreover, our model's calibrated prediction probabilities reduce the false discovery rate from 13.5% to 8.7% without missing any acute leukemia cases, thereby decreasing the number of unnecessary bone marrow aspirations based on peripheral blood smears. This study highlights the potential of AI-assisted diagnostics in hematological malignancies, illustrating how models trained on real-world data could enhance diagnostic accuracy and reduce invasive procedures.

This study introduces a parameter-efficient Hierarchical Spatial Temporal Network (HiSTN) specifically designed for the task of emotion classification using multi-channel electroencephalogram data. The network incorporates a graph hierarchy constructed from bottom-up at various abstraction levels, offering the dual advantages of enhanced task-relevant deep feature extraction and a lightweight design. The model's effectiveness is further amplified when used in conjunction with a proposed unique label smoothing method. Comprehensive benchmark experiments reveal that this combined approach yields high, balanced performance in terms of both quantitative and qualitative predictions. HiSTN, which has approximately 1,000 parameters, achieves mean F1 scores of 96.82% (valence) and 95.62% (arousal) in subject-dependent tests on the rarely-utilized 5-classification task problem from the DREAMER dataset. In the subject-independent settings, the same model yields mean F1 scores of 78.34% for valence and 81.59% for arousal. The adoption of the Sequential Top-2 Hit Rate (Seq2HR) metric highlights the significant enhancements in terms of the balance between model's quantitative and qualitative for predictions achieved through our approach when compared to training with regular one-hot labels. These improvements surpass 50% in subject-dependent tasks and 30% in subject-independent tasks. The study also includes relevant ablation studies and case explorations to further elucidate the workings of the proposed model and enhance its interpretability.

Visual Question Answering (VQA) is a challenging task of natural language processing (NLP) and computer vision (CV), attracting significant attention from researchers. English is a resource-rich language that has witnessed various developments in datasets and models for visual question answering. Visual question answering in other languages also would be developed for resources and models. In addition, there is no multilingual dataset targeting the visual content of a particular country with its own objects and cultural characteristics. To address the weakness, we provide the research community with a benchmark dataset named EVJVQA, including 33,000+ pairs of question-answer over three languages: Vietnamese, English, and Japanese, on approximately 5,000 images taken from Vietnam for evaluating multilingual VQA systems or models. EVJVQA is used as a benchmark dataset for the challenge of multilingual visual question answering at the 9th Workshop on Vietnamese Language and Speech Processing (VLSP 2022). This task attracted 62 participant teams from various universities and organizations. In this article, we present details of the organization of the challenge, an overview of the methods employed by shared-task participants, and the results. The highest performances are 0.4392 in F1-score and 0.4009 in BLUE on the private test set. The multilingual QA systems proposed by the top 2 teams use ViT for the pre-trained vision model and mT5 for the pre-trained language model, a powerful pre-trained language model based on the transformer architecture. EVJVQA is a challenging dataset that motivates NLP and CV researchers to further explore the multilingual models or systems for visual question answering systems. We released the challenge on the Codalab evaluation system for further research.

Generative retrieval (Wang et al., 2022; Tay et al., 2022) is a popular approach for end-to-end document retrieval that directly generates document identifiers given an input query. We introduce summarization-based document IDs, in which each document's ID is composed of an extractive summary or abstractive keyphrases generated by a language model, rather than an integer ID sequence or bags of n-grams as proposed in past work. We find that abstractive, content-based IDs (ACID) and an ID based on the first 30 tokens are very effective in direct comparisons with previous approaches to ID creation. We show that using ACID improves top-10 and top-20 recall by 15.6% and 14.4% (relative) respectively versus the cluster-based integer ID baseline on the MSMARCO 100k retrieval task, and 9.8% and 9.9% respectively on the Wikipedia-based NQ 100k retrieval task. Our results demonstrate the effectiveness of human-readable, natural-language IDs created through summarization for generative retrieval. We also observed that extractive IDs outperformed abstractive IDs on Wikipedia articles in NQ but not the snippets in MSMARCO, which suggests that document characteristics affect generative retrieval performance.

We present Brainstacks, a modular architecture for continual multi-domain fine-tuning of large language models that packages domain expertise as frozen adapter stacks composing additively on a shared frozen base at inference. Five interlocking components: (1) MoE-LoRA with Shazeer-style noisy top-2 routing across all seven transformer projections under QLoRA 4-bit quantization with rsLoRA scaling; (2) an inner loop performing residual boosting by freezing trained stacks and adding new ones; (3) an outer loop training sequential domain-specific stacks with curriculum-ordered dependencies; (4) null-space projection via randomized SVD constraining new stacks to subspaces orthogonal to prior directions, achieving zero forgetting in isolation; (5) an outcome-based sigmoid meta-router trained on empirically discovered domain-combination targets that selectively weights stacks, enabling cross-domain composition. Two boundary experiments: (6) PSN pretraining on a randomly initialized model; (7) per-domain RL (DPO/GRPO) validating compatibility with post-SFT alignment. Validated on TinyLlama-1.1B (4 domains, 9 stacks) and Gemma 3 12B IT (5 domains, 10 stacks), MoE-LoRA achieves 2.5x faster convergence than parameter-matched single LoRA, residual boosting breaks through the single-stack ceiling, and the routed system recovers generation quality destroyed by ungated stack accumulation. The central finding: the outcome-based router discovers that domain stacks encode transferable cognitive primitives (instruction-following clarity, numerical reasoning, procedural logic, chain-of-thought structure) rather than domain-specific knowledge, with medical prompts routing to chat+math stacks in 97% of cases despite zero medical data in those stacks.

Mechanism-targeted synthetic data is increasingly proposed as a way to steer pretraining toward desirable capabilities, but it remains unclear how such interventions should be evaluated. We study this question for in-context learning (ICL) under matched compute (iso-FLOPs) using Bi-Induct, a lightweight data rewrite that interleaves short directional copy snippets into a natural pretraining stream: forward-copy (induction), backward-copy (anti-induction, as a directional control), or a balanced mix. Across 0.13B-1B decoder-only models, we evaluate (i) few-shot performance on standard LM benchmarks and function-style ICL probes, (ii) head-level copy telemetry, and (iii) held-out perplexity as a guardrail. Bi-Induct reliably increases induction-head activity, but this does not translate into consistent improvements in few-shot generalization: on standard LM benchmarks, Bi-Induct is largely performance-neutral relative to natural-only training, while on function-style probes the 1B natural-only model performs best. Despite explicit backward-copy cues, anti-induction scores remain near zero across scales, revealing a strong forward/backward asymmetry. Targeted ablations show a sharper distinction: removing the top 2% induction heads per layer harms ICL more than matched random ablations, with the largest relative drop occurring in the natural-only models. This indicates that natural-only training produces more centralized, load-bearing induction circuitry, whereas Bi-Induct tends to create more distributed and redundant induction activity. Our main conclusion is that eliciting a mechanism is not the same as making it load-bearing. For data-centric foundation model design, this suggests that synthetic data interventions should be evaluated not only by signature amplification, but by whether they create causally necessary computation while preserving natural-data modeling quality.

The application of large vision-language models to computational pathology holds great promise for diagnostic assistants but faces a critical computational bottleneck: the gigapixel scale of Whole Slide Images (WSIs). A single WSI typically contains over 105 patches, creating sequence lengths that exceed the constraints of standard Transformer architectures. Existing solutions often resort to spatial sampling, which risks discarding diagnostically critical evidence. To address this, we propose TC-SSA (Token Compression via Semantic Slot Aggregation), a learnable token compression framework that aggregates patch features into a fixed number of semantic slots. A gated routing module assigns patches to slots using sparse Top-2 routing, followed by weighted aggregation, enabling global slide coverage under a strict token budget. The resulting representation retains diagnostically relevant information while reducing the number of visual tokens to 1.7% of the original sequence. On the SlideBench(TCGA), our model achieves 78.34% overall accuracy and 77.14% on the diagnosis subset, outperforming sampling-based baselines under comparable token budgets. The method also generalizes to MIL classification, reaching AUC of 95.83% on TCGA-BRCA, 98.27% on TCGA-NSCLC and 79.80% on PANDA. These results suggest that learnable semantic aggregation provides an effective trade-off between efficiency and diagnostic performance for gigapixel pathology reasoning.

Stock recommendation is vital to investment companies and investors. However, no single stock selection strategy will always win while analysts may not have enough time to check all S&P 500 stocks (the Standard & Poor's 500). In this paper, we propose a practical scheme that recommends stocks from S&P 500 using machine learning. Our basic idea is to buy and hold the top 20% stocks dynamically. First, we select representative stock indicators with good explanatory power. Secondly, we take five frequently used machine learning methods, including linear regression, ridge regression, stepwise regression, random forest and generalized boosted regression, to model stock indicators and quarterly log-return in a rolling window. Thirdly, we choose the model with the lowest Mean Square Error in each period to rank stocks. Finally, we test the selected stocks by conducting portfolio allocation methods such as equally weighted, mean-variance, and minimum-variance. Our empirical results show that the proposed scheme outperforms the long-only strategy on the S&P 500 index in terms of Sharpe ratio and cumulative returns. This work is fully open-sourced at https://github.com/AI4Finance-Foundation/Dynamic-Stock-Recommendation-Machine_Learning-Published-Paper-IEEE{GitHub}.

By 2028 most cybersecurity actions will be autonomous, with humans teleoperating. We present the first classification of autonomy levels in cybersecurity and introduce Cybersecurity AI (CAI), an open-source framework that democratizes advanced security testing through specialized AI agents. Through rigorous empirical evaluation, we demonstrate that CAI consistently outperforms state-of-the-art results in CTF benchmarks, solving challenges across diverse categories with significantly greater efficiency -up to 3,600x faster than humans in specific tasks and averaging 11x faster overall. CAI achieved first place among AI teams and secured a top-20 position worldwide in the "AI vs Human" CTF live Challenge, earning a monetary reward of $750. Based on our results, we argue against LLM-vendor claims about limited security capabilities. Beyond cybersecurity competitions, CAI demonstrates real-world effectiveness, reaching top-30 in Spain and top-500 worldwide on Hack The Box within a week, while dramatically reducing security testing costs by an average of 156x. Our framework transcends theoretical benchmarks by enabling non-professionals to discover significant security bugs (CVSS 4.3-7.5) at rates comparable to experts during bug bounty exercises. By combining modular agent design with seamless tool integration and human oversight (HITL), CAI addresses critical market gaps, offering organizations of all sizes access to AI-powered bug bounty security testing previously available only to well-resourced firms -thereby challenging the oligopolistic ecosystem currently dominated by major bug bounty platforms.

Modern code generation tools, utilizing AI models like Large Language Models (LLMs), have gained popularity for producing functional code. However, their usage presents security challenges, often resulting in insecure code merging into the code base. Evaluating the quality of generated code, especially its security, is crucial. While prior research explored various aspects of code generation, the focus on security has been limited, mostly examining code produced in controlled environments rather than real-world scenarios. To address this gap, we conducted an empirical study, analyzing code snippets generated by GitHub Copilot from GitHub projects. Our analysis identified 452 snippets generated by Copilot, revealing a high likelihood of security issues, with 32.8% of Python and 24.5% of JavaScript snippets affected. These issues span 38 different Common Weakness Enumeration (CWE) categories, including significant ones like CWE-330: Use of Insufficiently Random Values, CWE-78: OS Command Injection, and CWE-94: Improper Control of Generation of Code. Notably, eight CWEs are among the 2023 CWE Top-25, highlighting their severity. Our findings confirm that developers should be careful when adding code generated by Copilot and should also run appropriate security checks as they accept the suggested code. It also shows that practitioners should cultivate corresponding security awareness and skills.

In this paper, we explore the capabilities of LLMs in capturing lexical-semantic knowledge from WordNet on the example of the LLaMA-2-7b model and test it on multiple lexical semantic tasks. As the outcome of our experiments, we present TaxoLLaMA, the everything-in-one model, lightweight due to 4-bit quantization and LoRA. It achieves 11 SotA results, 4 top-2 results out of 16 tasks for the Taxonomy Enrichment, Hypernym Discovery, Taxonomy Construction, and Lexical Entailment tasks. Moreover, it demonstrates very strong zero-shot performance on Lexical Entailment and Taxonomy Construction with no fine-tuning. We also explore its hidden multilingual and domain adaptation capabilities with a little tuning or few-shot learning. All datasets, code, and model are available online at https://github.com/VityaVitalich/TaxoLLaMA

Sparsely-activated Mixture-of-experts (MoE) models allow the number of parameters to greatly increase while keeping the amount of computation for a given token or a given sample unchanged. However, a poor expert routing strategy (e.g. one resulting in load imbalance) can cause certain experts to be under-trained, leading to an expert being under or over-specialized. Prior work allocates a fixed number of experts to each token using a top-k function regardless of the relative importance of different tokens. To address this, we propose a heterogeneous mixture-of-experts employing an expert choice method. Instead of letting tokens select the top-k experts, we have experts selecting the top-k tokens. As a result, each token can be routed to a variable number of experts and each expert can have a fixed bucket size. We systematically study pre-training speedups using the same computational resources of the Switch Transformer top-1 and GShard top-2 gating of prior work and find that our method improves training convergence time by more than 2x. For the same computational cost, our method demonstrates higher performance in fine-tuning 11 selected tasks in the GLUE and SuperGLUE benchmarks. For a smaller activation cost, our method outperforms the T5 dense model in 7 out of the 11 tasks.

As autonomous AI agents transition from code completion tools to full-fledged teammates capable of opening pull requests (PRs) at scale, software maintainers face a new challenge: not just reviewing code, but managing complex interaction loops with non-human contributors. This paradigm shift raises a critical question: can we predict which agent-generated PRs will consume excessive review effort before any human interaction begins? Analyzing 33,707 agent-authored PRs from the AIDev dataset across 2,807 repositories, we uncover a striking two-regime behavioral pattern that fundamentally distinguishes autonomous agents from human developers. The first regime, representing 28.3 percent of all PRs, consists of instant merges (less than 1 minute), reflecting success on narrow automation tasks. The second regime involves iterative review cycles where agents frequently stall or abandon refinement (ghosting). We propose a Circuit Breaker triage model that predicts high-review-effort PRs (top 20 percent) at creation time using only static structural features. A LightGBM model achieves AUC 0.957 on a temporal split, while semantic text features (TF-IDF, CodeBERT) provide negligible predictive value. At a 20 percent review budget, the model intercepts 69 percent of total review effort, enabling zero-latency governance. Our findings challenge prevailing assumptions in AI-assisted code review: review burden is dictated by what agents touch, not what they say, highlighting the need for structural governance mechanisms in human-AI collaboration.

Billions of USD are invested in new artists and songs by the music industry every year. This research provides a new strategy for assessing the hit potential of songs, which can help record companies support their investment decisions. A number of models were developed that use both audio data, and a novel feature based on social media listening behaviour. The results show that models based on early adopter behaviour perform well when predicting top 20 dance hits.

Effectively training Large Language Models (LLMs) for complex, long-CoT reasoning is often bottlenecked by the need for massive high-quality reasoning data. Existing methods are either computationally expensive or fail to reliably distinguish high- from low-quality reasoning samples. To address this, we propose High-Entropy Sum (HES), a training-free metric that quantifies reasoning quality by summing only the entropy of the top (e.g., 0.5\%) highest-entropy tokens in each reasoning sample. We validate HES across three mainstream training paradigms: Supervised Fine-tuning (SFT), Rejection Fine-tuning (RFT), and Reinforcement Learning (RL), with extensive results demonstrating its consistent effectiveness and significantly reduced computational overhead. In SFT, training on the top 20\% HES-ranked data matches full-dataset performance, while using the lowest-HES data degrades it. In RFT, our HES-based training approach significantly outperforms baseline methods. In RL, HES-selected successful trajectories enable the model to learn strong reasoning patterns, significantly surpassing other compared methods. Our findings establish HES as a robust, training-free metric that enables a unified, effective, and efficient method for developing advanced reasoning in LLMs.

Modern astronomical surveys are producing datasets of unprecedented size and richness, increasing the potential for high-impact scientific discovery. This possibility, coupled with the challenge of exploring a large number of sources, has led to the development of novel machine-learning-based anomaly detection approaches, such as Astronomaly. For the first time, we test the scalability of Astronomaly by applying it to almost 4 million images of galaxies from the Dark Energy Camera Legacy Survey. We use a trained deep learning algorithm to learn useful representations of the images and pass these to the anomaly detection algorithm isolation forest, coupled with Astronomaly's active learning method, to discover interesting sources. We find that data selection criteria have a significant impact on the trade-off between finding rare sources such as strong lenses and introducing artefacts into the dataset. We demonstrate that active learning is required to identify the most interesting sources and reduce artefacts, while anomaly detection methods alone are insufficient. Using Astronomaly, we find 1635 anomalies among the top 2000 sources in the dataset after applying active learning, including eight strong gravitational lens candidates, 1609 galaxy merger candidates, and 18 previously unidentified sources exhibiting highly unusual morphology. Our results show that by leveraging the human-machine interface, Astronomaly is able to rapidly identify sources of scientific interest even in large datasets.

Recent work on training neural retrievers for open-domain question answering (OpenQA) has employed both supervised and unsupervised approaches. However, it remains unclear how unsupervised and supervised methods can be used most effectively for neural retrievers. In this work, we systematically study retriever pre-training. We first propose an approach of unsupervised pre-training with the Inverse Cloze Task and masked salient spans, followed by supervised finetuning using question-context pairs. This approach leads to absolute gains of 2+ points over the previous best result in the top-20 retrieval accuracy on Natural Questions and TriviaQA datasets. We also explore two approaches for end-to-end supervised training of the reader and retriever components in OpenQA models. In the first approach, the reader considers each retrieved document separately while in the second approach, the reader considers all the retrieved documents together. Our experiments demonstrate the effectiveness of these approaches as we obtain new state-of-the-art results. On the Natural Questions dataset, we obtain a top-20 retrieval accuracy of 84, an improvement of 5 points over the recent DPR model. In addition, we achieve good results on answer extraction, outperforming recent models like REALM and RAG by 3+ points. We further scale up end-to-end training to large models and show consistent gains in performance over smaller models.

Multimedia generation approaches occupy a prominent place in artificial intelligence research. Text-to-image models achieved high-quality results over the last few years. However, video synthesis methods recently started to develop. This paper presents a new two-stage latent diffusion text-to-video generation architecture based on the text-to-image diffusion model. The first stage concerns keyframes synthesis to figure the storyline of a video, while the second one is devoted to interpolation frames generation to make movements of the scene and objects smooth. We compare several temporal conditioning approaches for keyframes generation. The results show the advantage of using separate temporal blocks over temporal layers in terms of metrics reflecting video generation quality aspects and human preference. The design of our interpolation model significantly reduces computational costs compared to other masked frame interpolation approaches. Furthermore, we evaluate different configurations of MoVQ-based video decoding scheme to improve consistency and achieve higher PSNR, SSIM, MSE, and LPIPS scores. Finally, we compare our pipeline with existing solutions and achieve top-2 scores overall and top-1 among open-source solutions: CLIPSIM = 0.2976 and FVD = 433.054. Project page: https://ai-forever.github.io/kandinsky-video/

Mixture-of-Experts (MoE) models scale more effectively than dense models due to sparse computation through expert routing, selectively activating only a small subset of expert modules. However, sparse computation challenges traditional training practices, as discrete expert routing hinders standard backpropagation and thus gradient-based optimization, which are the cornerstone of deep learning. To better pursue the scaling power of MoE, we introduce GRIN (GRadient-INformed MoE training), which incorporates sparse gradient estimation for expert routing and configures model parallelism to avoid token dropping. Applying GRIN to autoregressive language modeling, we develop a top-2 16times3.8B MoE model. Our model, with only 6.6B activated parameters, outperforms a 7B dense model and matches the performance of a 14B dense model trained on the same data. Extensive evaluations across diverse tasks demonstrate the potential of GRIN to significantly enhance MoE efficacy, achieving 79.4 on MMLU, 83.7 on HellaSwag, 74.4 on HumanEval, and 58.9 on MATH.

The rapid growth of user-generated (video) content (UGC) has driven increased demand for research on no-reference (NR) perceptual video quality assessment (VQA). NR-VQA is a key component for large-scale video quality monitoring in social media and streaming applications where a pristine reference is not available. This paper proposes a novel NR-VQA model based on spatio-temporal fragmentation driven by inter-frame variations. By leveraging these inter-frame differences, the model progressively analyses quality-sensitive regions at multiple levels: frames, patches, and fragmented frames. It integrates frames, fragmented residuals, and fragmented frames aligned with residuals to effectively capture global and local information. The model extracts both 2D and 3D features in order to characterize these spatio-temporal variations. Experiments conducted on five UGC datasets and against state-of-the-art models ranked our proposed method among the top 2 in terms of average rank correlation (DIVA-VQA-L: 0.898 and DIVA-VQA-B: 0.886). The improved performance is offered at a low runtime complexity, with DIVA-VQA-B ranked top and DIVA-VQA-L third on average compared to the fastest existing NR-VQA method. Code and models are publicly available at: https://github.com/xinyiW915/DIVA-VQA.

Scaling large language models (LLMs) significantly improves performance but comes with prohibitive computational costs. Mixture-of-Experts (MoE) models offer an efficient alternative, increasing capacity without a proportional rise in compute requirements. However, training MoE models from scratch poses challenges like overfitting and routing instability. We present an efficient training recipe leveraging pre-trained dense checkpoints, training an 8-Expert Top-2 MoE model from Llama 3-8B with less than 1% of typical pre-training compute. Our approach enhances downstream performance on academic benchmarks, achieving a 2% improvement in 0-shot accuracy on MMLU, while reaching a Model FLOPs Utilization (MFU) of 46.8% during training using our framework. We also integrate online upcycling in NeMo for seamless use of pre-trained weights, enabling cost-effective development of high-capacity MoE models.

Aggregate metacognitive quality scores mask within-model variation across MMLU benchmark domains. We administered 1,500 MMLU items (250 per domain, under an a priori six-domain grouping) to 33 frontier LLMs from eight model families and computed Type-2 AUROC per model-domain cell using verbalized confidence (0-100). Total observations: 47,151. Every model with above-chance aggregate monitoring showed non-trivial domain-level variation. Applied/Professional knowledge was reliably the easiest benchmark domain to monitor (mean AUROC = .742, ranked top-2 in 21 of 33 models); Formal Reasoning and Natural Science were reliably the hardest (one of the two ranked bottom-2 in 27 of 33 models). The three middle domains were statistically indistinguishable (Kendall's W = .164). A subject-level coherence analysis (within-domain similarity ratio = 0.95) confirms the six-domain grouping is a pragmatic benchmark taxonomy, not a validated latent construct. Within-family profile-shape clustering is significant for Anthropic, Google-Gemini, and Qwen (permutation p < .0001) but not DeepSeek, Google-Gemma, or OpenAI. Gemma 4 31B showed a +.202 AUROC improvement over Gemma 3 27B. Three models classified Invalid on binary KEEP/WITHDRAW probes produced normal profiles under verbalized confidence, confirming probe-format specificity. Bootstrap 95% CIs on 198 cells have median width .199. Split-half aggregate stability r = .893; profile-level split-half is weaker (grand median r = .184). These results show stable benchmark-domain variation obscured by aggregate metrics, and support benchmark-stage domain screening as a step before deployment in specific application areas.

LLM unlearning has emerged as a promising approach, aiming to enable models to forget hazardous/undesired knowledge at low cost while preserving as much model utility as possible. Among existing techniques, the most straightforward method is performing Gradient Ascent (GA) w.r.t. the forget data, thereby forcing the model to unlearn the forget dataset. However, GA suffers from severe instability, as it drives updates in a divergent direction, often resulting in drastically degraded model utility. To address this issue, we propose Smoothed Gradient Ascent (SGA). SGA combines the forget data with multiple constructed normal data through a tunable smoothing rate. Intuitively, this extends GA from learning solely on the forget data to jointly learning across both forget and normal data, enabling more stable unlearning while better preserving model utility. Theoretically, we provide the theoretical guidance on the selection of the optimal smoothing rate. Empirically, we evaluate SGA on three benchmarks: TOFU, Harry Potter, and MUSE-NEWS. Experimental results demonstrate that SGA consistently outperforms the original Gradient Ascent (GA) method across all metrics and achieves top-2 performance among all baseline methods on several key metrics.

In this paper, we introduce a novel dynamic expert selection framework for Mixture of Experts (MoE) models, aiming to enhance computational efficiency and model performance by adjusting the number of activated experts based on input difficulty. Unlike traditional MoE approaches that rely on fixed Top-K routing, which activates a predetermined number of experts regardless of the input's complexity, our method dynamically selects experts based on the confidence level in expert selection for each input. This allows for a more efficient utilization of computational resources, activating more experts for complex tasks requiring advanced reasoning and fewer for simpler tasks. Through extensive evaluations, our dynamic routing method demonstrates substantial improvements over conventional Top-2 routing across various benchmarks, achieving an average improvement of 0.7% with less than 90% activated parameters. Further analysis shows our model dispatches more experts to tasks requiring complex reasoning skills, like BBH, confirming its ability to dynamically allocate computational resources in alignment with the input's complexity. Our findings also highlight a variation in the number of experts needed across different layers of the transformer model, offering insights into the potential for designing heterogeneous MoE frameworks. The code and models are available at https://github.com/ZhenweiAn/Dynamic_MoE.

Molecular structure elucidation from spectra is a foundational problem in chemistry, with profound implications for compound identification, synthesis, and drug development. Traditional methods rely heavily on expert interpretation and lack scalability. Pioneering machine learning methods have introduced retrieval-based strategies, but their reliance on finite libraries limits generalization to novel molecules. Generative models offer a promising alternative, yet most adopt autoregressive SMILES-based architectures that overlook 3D geometry and struggle to integrate diverse spectral modalities. In this work, we present DiffSpectra, a generative framework that directly infers both 2D and 3D molecular structures from multi-modal spectral data using diffusion models. DiffSpectra formulates structure elucidation as a conditional generation process. Its denoising network is parameterized by Diffusion Molecule Transformer, an SE(3)-equivariant architecture that integrates topological and geometric information. Conditioning is provided by SpecFormer, a transformer-based spectral encoder that captures intra- and inter-spectral dependencies from multi-modal spectra. Extensive experiments demonstrate that DiffSpectra achieves high accuracy in structure elucidation, recovering exact structures with 16.01% top-1 accuracy and 96.86% top-20 accuracy through sampling. The model benefits significantly from 3D geometric modeling, SpecFormer pre-training, and multi-modal conditioning. These results highlight the effectiveness of spectrum-conditioned diffusion modeling in addressing the challenge of molecular structure elucidation. To our knowledge, DiffSpectra is the first framework to unify multi-modal spectral reasoning and joint 2D/3D generative modeling for de novo molecular structure elucidation.

Large Language Models (LLMs) have demonstrated impressive performance on Natural Language Processing (NLP) tasks, such as Question Answering, Summarization, and Classification. The use of LLMs as evaluators, that can rank or score the output of other models (usually LLMs) has become increasingly popular, due to the limitations of current evaluation techniques including the lack of appropriate benchmarks, metrics, cost, and access to human annotators. While LLMs are capable of handling approximately 100 languages, the majority of languages beyond the top 20 lack systematic evaluation across various tasks, metrics, and benchmarks. This creates an urgent need to scale up multilingual evaluation to ensure a precise understanding of LLM performance across diverse languages. LLM-based evaluators seem like the perfect solution to this problem, as they do not require human annotators, human-created references, or benchmarks and can theoretically be used to evaluate any language covered by the LLM. In this paper, we investigate whether LLM-based evaluators can help scale up multilingual evaluation. Specifically, we calibrate LLM-based evaluation against 20k human judgments of five metrics across three text-generation tasks in eight languages. Our findings indicate that LLM-based evaluators may exhibit bias towards higher scores and should be used with caution and should always be calibrated with a dataset of native speaker judgments, particularly in low-resource and non-Latin script languages.

Recent studies have shown that maintaining a consistent response style by human experts and enhancing data quality in training sets can significantly improve the performance of fine-tuned Large Language Models (LLMs) while reducing the number of training examples needed. However, the precise definition of style and the relationship between style, data quality, and LLM performance remains unclear. This research decomposes response style into presentation and composition styles and finds that, among training data of similar quality, those with higher style consistency lead to better LLM performance. Inspired by this, we introduce Style Consistency-Aware Response Ranking (SCAR), which automatically prioritizes instruction-response pairs in the training set based on their response stylistic consistency. By selecting the most style-consistent examples, ranging from the top 25% to 0.7% of the full dataset, the fine-tuned LLMs can match or even surpass the performance of models trained on the entire dataset in coding and open-ended question-answering benchmarks. Code and data are available at https://github.com/zhuang-li/SCAR .

LLM benchmark labels are frozen at release and silently propagated into downstream benchmarks, errors and all. We introduce an Item Response Theory-based indicator that surfaces likely mislabels at 95% precision in the top 200 examples across seven preference and multiple-choice benchmarks using responses from 114 models, outperforming a supervised classifier. We trace these errors to mechanical labeling heuristics, upstream annotation mistakes inherited unchanged from source datasets, and fundamentally ambiguous items without a defensible single label. The same model fit reveals that reward models specialize in stylistic preference rather than factual knowledge, and identifies one frontier reward model that agrees with detected mislabels at 78% accuracy versus 38% for its peers, consistent with benchmark contamination or benchmark-specific over-optimization.

This study presents a comparative analysis of 55 Wikipedia language editions employing a citation index alongside a synthetic quality measure. Specifically, we identified the most significant Wikipedia articles within distinct topical areas, selecting the top 10, top 25, and top 100 most cited articles in each topic and language version. This index was built on the basis of wikilinks between Wikipedia articles in each language version and in order to do that we processed 6.6 billion page-to-page link records. Next, we used a quality score for each Wikipedia article - a synthetic measure scaled from 0 to 100. This approach enabled quality comparison of Wikipedia articles even between language versions with different quality grading schemes. Our results highlight disparities among Wikipedia language editions, revealing strengths and gaps in content coverage and quality across topics.

Large Language Models (LLMs) have demonstrated significant capabilities in understanding and analyzing code for security vulnerabilities, such as Common Weakness Enumerations (CWEs). However, their reliance on cloud infrastructure and substantial computational requirements pose challenges for analyzing sensitive or proprietary codebases due to privacy concerns and inference costs. This work explores the potential of Small Language Models (SLMs) as a viable alternative for accurate, on-premise vulnerability detection. We investigated whether a 350-million parameter pre-trained code model (codegen-mono) could be effectively fine-tuned to detect the MITRE Top 25 CWEs specifically within Python code. To facilitate this, we developed a targeted dataset of 500 examples using a semi-supervised approach involving LLM-driven synthetic data generation coupled with meticulous human review. Initial tests confirmed that the base codegen-mono model completely failed to identify CWEs in our samples. However, after applying instruction-following fine-tuning, the specialized SLM achieved remarkable performance on our test set, yielding approximately 99% accuracy, 98.08% precision, 100% recall, and a 99.04% F1-score. These results strongly suggest that fine-tuned SLMs can serve as highly accurate and efficient tools for CWE detection, offering a practical and privacy-preserving solution for integrating advanced security analysis directly into development workflows.

Recent generative reasoning breakthroughs have transformed how large language models (LLMs) tackle complex problems by dynamically retrieving and refining information while generating coherent, multi-step thought processes. Techniques such as inference-time scaling, reinforcement learning, supervised fine-tuning, and distillation have been successfully applied to models like DeepSeek-R1, OpenAI's o1 & o3, GPT-4o, Qwen-32B, and various Llama variants, resulting in enhanced reasoning capabilities. In this paper, we provide a comprehensive analysis of the top 27 LLM models released between 2023 and 2025 (including models such as Mistral AI Small 3 24B, DeepSeek-R1, Search-o1, QwQ-32B, and phi-4). Then, we present an extensive overview of training methodologies that spans general training approaches, mixture-of-experts (MoE) and architectural innovations, retrieval-augmented generation (RAG), chain-of-thought and self-improvement techniques, as well as test-time compute scaling, distillation, and reinforcement learning (RL) methods. Finally, we discuss the key challenges in advancing LLM capabilities, including improving multi-step reasoning without human supervision, overcoming limitations in chained tasks, balancing structured prompts with flexibility, and enhancing long-context retrieval and external tool integration.

Diffusion language models (DLMs) have recently emerged as an alternative to autoregressive approaches, offering parallel sequence generation and flexible token orders. However, their inference remains slower than that of autoregressive models, primarily due to the cost of bidirectional attention and the large number of refinement steps required for high quality outputs. In this work, we highlight and leverage an overlooked property of DLMs early answer convergence: in many cases, the correct answer can be internally identified by half steps before the final decoding step, both under semi-autoregressive and random remasking schedules. For example, on GSM8K and MMLU, up to 97% and 99% of instances, respectively, can be decoded correctly using only half of the refinement steps. Building on this observation, we introduce Prophet, a training-free fast decoding paradigm that enables early commit decoding. Specifically, Prophet dynamically decides whether to continue refinement or to go "all-in" (i.e., decode all remaining tokens in one step), using the confidence gap between the top-2 prediction candidates as the criterion. It integrates seamlessly into existing DLM implementations, incurs negligible overhead, and requires no additional training. Empirical evaluations of LLaDA-8B and Dream-7B across multiple tasks show that Prophet reduces the number of decoding steps by up to 3.4x while preserving high generation quality. These results recast DLM decoding as a problem of when to stop sampling, and demonstrate that early decode convergence provides a simple yet powerful mechanism for accelerating DLM inference, complementary to existing speedup techniques. Our code is publicly available at https://github.com/pixeli99/Prophet.

Sparse mixture-of-experts (MoE) language models route each token to multiple experts, suggesting a geometric account of their benefit: co-selected experts should contribute distinct representation directions. Existing evidence often conflates route coherence, candidate quality, and candidate-by-context interaction. We distinguish these quantities using an Expert Subspace Separation Index (ESSI), matched-route residuals, and a prefix-controlled 2times2 factorial; frozen-route interventions and a controlled Top-k study assess functional value. Three paired contrasts organize the findings. First, across six MoE architectures, expert subspaces overlap substantially, yet actual routes explain token representations better than matched alternatives. Second, across the 39 factorial cells in OLMoE, Mixtral, and DeepSeek, the selected candidate explains more of the residual representation than the strongest unselected rival in every cell, yet the actual prefix narrows this advantage throughout: all interactions are negative, and every 95% confidence interval lies below zero. Third, this geometric narrowing does not imply functional redundancy: adding later experts improves next-token prediction in 24 of 39 frozen-route comparisons, while the other 15 estimates are inconclusive; a controlled training study also favors Top-2 over Top-1 in all three seeds. We call this joint pattern coherent overlap: routing selects token-relevant experts from a shared geometric neighborhood, while useful multi-expert computation persists without disjoint linear coverage. Separating these quantities clarifies why geometric similarity alone cannot determine redundancy or pruning value.

Ultrasound fetal imaging is beneficial to support prenatal development because it is affordable and non-intrusive. Nevertheless, fetal plane classification (FPC) remains challenging and time-consuming for obstetricians since it depends on nuanced clinical aspects, which increases the difficulty in identifying relevant features of the fetal anatomy. Thus, to assist with its accurate feature extraction, a lightweight artificial intelligence architecture leveraging convolutional neural networks and attention mechanisms is proposed to classify the largest benchmark ultrasound dataset. The approach fine-tunes from lightweight EfficientNet feature extraction backbones pre-trained on the ImageNet1k. to classify key fetal planes such as the brain, femur, thorax, cervix, and abdomen. Our methodology incorporates the attention mechanism to refine features and 3-layer perceptrons for classification, achieving superior performance with the highest Top-1 accuracy of 96.25%, Top-2 accuracy of 99.80% and F1-Score of 0.9576. Importantly, the model has 40x fewer trainable parameters than existing benchmark ensemble or transformer pipelines, facilitating easy deployment on edge devices to help clinical practitioners with real-time FPC. The findings are also interpreted using GradCAM to carry out clinical correlation to aid doctors with diagnostics and improve treatment plans for expectant mothers.

We present rStar-Math to demonstrate that small language models (SLMs) can rival or even surpass the math reasoning capability of OpenAI o1, without distillation from superior models. rStar-Math achieves this by exercising "deep thinking" through Monte Carlo Tree Search (MCTS), where a math policy SLM performs test-time search guided by an SLM-based process reward model. rStar-Math introduces three innovations to tackle the challenges in training the two SLMs: (1) a novel code-augmented CoT data sythesis method, which performs extensive MCTS rollouts to generate step-by-step verified reasoning trajectories used to train the policy SLM; (2) a novel process reward model training method that avoids na\"ive step-level score annotation, yielding a more effective process preference model (PPM); (3) a self-evolution recipe in which the policy SLM and PPM are built from scratch and iteratively evolved to improve reasoning capabilities. Through 4 rounds of self-evolution with millions of synthesized solutions for 747k math problems, rStar-Math boosts SLMs' math reasoning to state-of-the-art levels. On the MATH benchmark, it improves Qwen2.5-Math-7B from 58.8% to 90.0% and Phi3-mini-3.8B from 41.4% to 86.4%, surpassing o1-preview by +4.5% and +0.9%. On the USA Math Olympiad (AIME), rStar-Math solves an average of 53.3% (8/15) of problems, ranking among the top 20% the brightest high school math students. Code and data will be available at https://github.com/microsoft/rStar.

Large language models (LLMs) have revolutionized natural language processing (NLP), yet open-source multilingual LLMs remain scarce, with existing models often limited in language coverage. Such models typically prioritize well-resourced languages, while widely spoken but under-resourced languages are often overlooked. To address this disparity, we introduce Babel, an open multilingual LLM that covers the top 25 languages by number of speakers, supports over 90% of the global population, and includes many languages neglected by other open multilingual LLMs. Unlike traditional continue pretraining approaches, Babel expands its parameter count through a layer extension technique that elevates Babel's performance ceiling. We introduce two variants: Babel-9B, designed for efficient inference and fine-tuning, and Babel-83B, which sets a new standard for open multilingual LLMs. Extensive evaluations on multilingual tasks demonstrate its superior performance compared to open LLMs of comparable size. In addition, using open-source supervised fine-tuning datasets, Babel achieves remarkable performance, with Babel-9B-Chat leading among 10B-sized LLMs and Babel-83B-Chat setting a new standard for multilingual tasks, reaching the same level of commercial models.

How do large language models understand moral dimensions compared to humans? This first large-scale Bayesian evaluation of market-leading language models provides the answer. In contrast to prior work using deterministic ground truth (majority or inclusion rules), we model annotator disagreements to capture both aleatoric uncertainty (inherent human disagreement) and epistemic uncertainty (model domain sensitivity). We evaluate top language models (Claude Sonnet 4, DeepSeek-V3, Llama 4 Maverick) across 250K+ annotations from ~700 annotators on 100K+ texts spanning social media, news, and forums. Our GPU-optimized Bayesian framework processed 1M+ model queries, revealing that AI models typically rank among the top 25\% of human annotators, achieving much better-than-average balanced accuracy. Importantly, we find that AI produces far fewer false negatives than humans, highlighting their more sensitive moral detection capabilities.

Open-domain question answering relies on efficient passage retrieval to select candidate contexts, where traditional sparse vector space models, such as TF-IDF or BM25, are the de facto method. In this work, we show that retrieval can be practically implemented using dense representations alone, where embeddings are learned from a small number of questions and passages by a simple dual-encoder framework. When evaluated on a wide range of open-domain QA datasets, our dense retriever outperforms a strong Lucene-BM25 system largely by 9%-19% absolute in terms of top-20 passage retrieval accuracy, and helps our end-to-end QA system establish new state-of-the-art on multiple open-domain QA benchmarks.

Mathematical modeling is a cornerstone of scientific discovery and engineering practice, enabling the translation of real-world problems into formal systems across domains such as physics, biology, and economics. Unlike mathematical reasoning, which assumes a predefined formulation, modeling requires open-ended problem analysis, abstraction, and principled formalization. While Large Language Models (LLMs) have shown strong reasoning capabilities, they fall short in rigorous model construction, limiting their utility in real-world problem-solving. To this end, we formalize the task of LLM-powered real-world mathematical modeling, where agents must analyze problems, construct domain-appropriate formulations, and generate complete end-to-end solutions. We introduce MM-Bench, a curated benchmark of 111 problems from the Mathematical Contest in Modeling (MCM/ICM), spanning the years 2000 to 2025 and across ten diverse domains such as physics, biology, and economics. To tackle this task, we propose MM-Agent, an expert-inspired framework that decomposes mathematical modeling into four stages: open-ended problem analysis, structured model formulation, computational problem solving, and report generation. Experiments on MM-Bench show that MM-Agent significantly outperforms baseline agents, achieving an 11.88\% improvement over human expert solutions while requiring only 15 minutes and \$0.88 per task using GPT-4o. Furthermore, under official MCM/ICM protocols, MM-Agent assisted two undergraduate teams in winning the Finalist Award (top 2.0\% among 27,456 teams) in MCM/ICM 2025, demonstrating its practical effectiveness as a modeling copilot. Our code is available at https://github.com/usail-hkust/LLM-MM-Agent

PubTator 3.0 (https://www.ncbi.nlm.nih.gov/research/pubtator3/) is a biomedical literature resource using state-of-the-art AI techniques to offer semantic and relation searches for key concepts like proteins, genetic variants, diseases, and chemicals. It currently provides over one billion entity and relation annotations across approximately 36 million PubMed abstracts and 6 million full-text articles from the PMC open access subset, updated weekly. PubTator 3.0's online interface and API utilize these precomputed entity relations and synonyms to provide advanced search capabilities and enable large-scale analyses, streamlining many complex information needs. We showcase the retrieval quality of PubTator 3.0 using a series of entity pair queries, demonstrating that PubTator 3.0 retrieves a greater number of articles than either PubMed or Google Scholar, with higher precision in the top 20 results. We further show that integrating ChatGPT (GPT-4) with PubTator APIs dramatically improves the factuality and verifiability of its responses. In summary, PubTator 3.0 offers a comprehensive set of features and tools that allow researchers to navigate the ever-expanding wealth of biomedical literature, expediting research and unlocking valuable insights for scientific discovery.

Large language models (LLMs) have transformed code generation, yet their application in hardware design produces gate counts 38\%--1075\% higher than human designs. We present CircuitMind, a multi-agent framework that achieves human-competitive efficiency through three key innovations: syntax locking (constraining generation to basic logic gates), retrieval-augmented generation (enabling knowledge-driven design), and dual-reward optimization (balancing correctness with efficiency). To evaluate our approach, we introduce TC-Bench, the first gate-level benchmark harnessing collective intelligence from the TuringComplete ecosystem -- a competitive circuit design platform with hundreds of thousands of players. Experiments show CircuitMind enables 55.6\% of model implementations to match or exceed top-tier human experts in composite efficiency metrics. Most remarkably, our framework elevates the 14B Phi-4 model to outperform both GPT-4o mini and Gemini 2.0 Flash, achieving efficiency comparable to the top 25\% of human experts without requiring specialized training. These innovations establish a new paradigm for hardware optimization where collaborative AI systems leverage collective human expertise to achieve optimal circuit designs. Our model, data, and code are open-source at https://github.com/BUAA-CLab/CircuitMind.

In this work, we introduce back-training, an alternative to self-training for unsupervised domain adaptation (UDA) from source to target domain. While self-training generates synthetic training data where natural inputs are aligned with noisy outputs, back-training results in natural outputs aligned with noisy inputs. This significantly reduces the gap between the target domain and synthetic data distribution, and reduces model overfitting to the source domain. We run UDA experiments on question generation and passage retrieval from the Natural Questions domain to machine learning and biomedical domains. We find that back-training vastly outperforms self-training by a mean improvement of 7.8 BLEU-4 points on generation, and 17.6\% top-20 retrieval accuracy across both domains. We further propose consistency filters to remove low-quality synthetic data before training. We also release a new domain-adaptation dataset- MLQuestions containing 35K unaligned questions, 50K unaligned passages, and 3K aligned question-passage pairs.

Hallucinations remain a major obstacle for large language models (LLMs), especially in safety-critical domains. We present HALT (Hallucination Assessment via Log-probs as Time series), a lightweight hallucination detector that leverages only the top-20 token log-probabilities from LLM generations as a time series. HALT uses a gated recurrent unit model combined with entropy-based features to learn model calibration bias, providing an extremely efficient alternative to large encoders. Unlike white-box approaches, HALT does not require access to hidden states or attention maps, relying only on output log-probabilities. Unlike black-box approaches, it operates on log-probs rather than surface-form text, which enables stronger domain generalization and compatibility with proprietary LLMs without requiring access to internal weights. To benchmark performance, we introduce HUB (Hallucination detection Unified Benchmark), which consolidates prior datasets into ten capabilities covering both reasoning tasks (Algorithmic, Commonsense, Mathematical, Symbolic, Code Generation) and general purpose skills (Chat, Data-to-Text, Question Answering, Summarization, World Knowledge). While being 30x smaller, HALT outperforms Lettuce, a fine-tuned modernBERT-base encoder, achieving a 60x speedup gain on HUB. HALT and HUB together establish an effective framework for hallucination detection across diverse LLM capabilities.

Capture-the-Flag (CTF) competitions are crucial for cybersecurity education and training. As large language models (LLMs) evolve, there is increasing interest in their ability to automate CTF challenge solving. For example, DARPA has organized the AIxCC competition since 2023 to advance AI-powered automated offense and defense. However, this demands a combination of multiple abilities, from knowledge to reasoning and further to actions. In this paper, we highlight the importance of technical knowledge in solving CTF problems and deliberately construct a focused benchmark, CTFKnow, with 3,992 questions to measure LLMs' performance in this core aspect. Our study offers a focused and innovative measurement of LLMs' capability in understanding CTF knowledge and applying it to solve CTF challenges. Our key findings reveal that while LLMs possess substantial technical knowledge, they falter in accurately applying this knowledge to specific scenarios and adapting their strategies based on feedback from the CTF environment. Based on insights derived from this measurement study, we propose CTFAgent, a novel LLM-driven framework for advancing CTF problem-solving. CTFAgent introduces two new modules: two-stage Retrieval Augmented Generation (RAG) and interactive Environmental Augmentation, which enhance LLMs' technical knowledge and vulnerability exploitation on CTF, respectively. Our experimental results show that, on two popular CTF datasets, CTFAgent both achieves over 80% performance improvement. Moreover, in the recent picoCTF2024 hosted by CMU, CTFAgent ranked in the top 23.6% of nearly 7,000 participating teams. This reflects the benefit of our measurement study and the potential of our framework in advancing LLMs' capabilities in CTF problem-solving.

Large language models (LLMs) have shown promise in transforming machine learning research, yet their capability to faithfully implement novel ideas from recent research papers-ideas unseen during pretraining-remains unclear. We introduce ResearchCodeBench, a benchmark of 212 coding challenges that evaluates LLMs' ability to translate cutting-edge ML contributions from top 2024-2025 research papers into executable code. We assessed 30+ proprietary and open-source LLMs, finding that even the best models correctly implement less than 40% of the code. We find Gemini-2.5-Pro-Preview to perform best at 37.3% success rate, with O3 (High) and O4-mini (High) following behind at 32.3% and 30.8% respectively. We present empirical findings on performance comparison, contamination, and error patterns. By providing a rigorous and community-driven evaluation platform, ResearchCodeBench enables continuous understanding and advancement of LLM-driven innovation in research code generation.

Not a day goes by without hearing about the impressive feats of large language models (LLMs), and equally, not a day passes without hearing about their challenges. LLMs are notoriously vulnerable to biases in their dataset, leading to issues such as toxicity. While domain-adaptive training has been employed to mitigate these issues, these techniques often address all model parameters indiscriminately during the repair process, resulting in poor repair quality and reduced model versatility. In this paper, we introduce a novel dynamic slicing-based intent-aware LLM repair strategy, IRepair. This approach selectively targets the most error-prone sections of the model for repair. Specifically, we propose dynamically slicing the model's most sensitive layers that require immediate attention, concentrating repair efforts on those areas. This method enables more effective repairs with potentially less impact on the model's overall performance by altering a smaller portion of the model. We evaluated our technique on three models from the GPT2 and GPT-Neo families, with parameters ranging from 800M to 1.6B, in a toxicity mitigation setup. Our results show that IRepair repairs errors 43.6% more effectively while causing 46% less disruption to general performance compared to the closest baseline, direct preference optimization. Our empirical analysis also reveals that errors are more concentrated in a smaller section of the model, with the top 20% of layers exhibiting 773% more error density than the remaining 80\%. This highlights the need for selective repair. Additionally, we demonstrate that a dynamic selection approach is essential for addressing errors dispersed throughout the model, ensuring a robust and efficient repair.

The drug development pipeline for a new compound can last 10-20 years and cost over 10 billion. Drug repurposing offers a more time- and cost-effective alternative. Computational approaches based on biomedical knowledge graph representations have recently yielded new drug repurposing hypotheses. In this study, we present a novel, disease-specific hypergraph representation learning technique to derive contextual embeddings of biological pathways of various lengths but that all start at any given drug and all end at the disease of interest. Further, we extend this method to multi-disease hypergraphs. To determine the repurposing potential of each of the 1,522 drugs, we derive drug-specific distributions of cosine similarity values and ultimately consider the median for ranking. Cosine similarity values are computed between (1) all biological pathways starting at the considered drug and ending at the disease of interest and (2) all biological pathways starting at drugs currently prescribed against that disease and ending at the disease of interest. We illustrate our approach with Alzheimer's disease (AD) and two of its risk factors: hypertension (HTN) and type 2 diabetes (T2D). We compare each drug's rank across four hypergraph settings (single- or multi-disease): AD only, AD + HTN, AD + T2D, and AD + HTN + T2D. Notably, our framework led to the identification of two promising drugs whose repurposing potential was significantly higher in hypergraphs combining two diseases: dapagliflozin (antidiabetic; moved up, from top 32% to top 7%, across all considered drugs) and debrisoquine (antihypertensive; moved up, from top 76% to top 23%). Our approach serves as a hypothesis generation tool, to be paired with a validation pipeline relying on laboratory experiments and semi-automated parsing of the biomedical literature.

Human evaluation is increasingly critical for assessing large language models, capturing linguistic nuances, and reflecting user preferences more accurately than traditional automated metrics. However, the resource-intensive nature of this type of annotation process poses significant challenges. The key question driving our work: "is it feasible to minimize human-in-the-loop feedback by prioritizing data instances which most effectively distinguish between models?" We evaluate several metric-based methods and find that these metrics enhance the efficiency of human evaluations by minimizing the number of required annotations, thus saving time and cost, while ensuring a robust performance evaluation. We show that our method is effective across widely used model families, reducing instances of indecisive (or "tie") outcomes by up to 54% compared to a random sample when focusing on the top-20 percentile of prioritized instances. This potential reduction in required human effort positions our approach as a valuable strategy in future large language model evaluations.

We propose a simple and effective re-ranking method for improving passage retrieval in open question answering. The re-ranker re-scores retrieved passages with a zero-shot question generation model, which uses a pre-trained language model to compute the probability of the input question conditioned on a retrieved passage. This approach can be applied on top of any retrieval method (e.g. neural or keyword-based), does not require any domain- or task-specific training (and therefore is expected to generalize better to data distribution shifts), and provides rich cross-attention between query and passage (i.e. it must explain every token in the question). When evaluated on a number of open-domain retrieval datasets, our re-ranker improves strong unsupervised retrieval models by 6%-18% absolute and strong supervised models by up to 12% in terms of top-20 passage retrieval accuracy. We also obtain new state-of-the-art results on full open-domain question answering by simply adding the new re-ranker to existing models with no further changes.

In 3D Visual Question Answering (3D VQA), the scarcity of fully annotated data and limited visual content diversity hampers the generalization to novel scenes and 3D concepts (e.g., only around 800 scenes are utilized in ScanQA and SQA dataset). Current approaches resort supplement 3D reasoning with 2D information. However, these methods face challenges: either they use top-down 2D views that introduce overly complex and sometimes question-irrelevant visual clues, or they rely on globally aggregated scene/image-level representations from 2D VLMs, losing the fine-grained vision-language correlations. To overcome these limitations, our approach utilizes question-conditional 2D view selection procedure, pinpointing semantically relevant 2D inputs for crucial visual clues. We then integrate this 2D knowledge into the 3D-VQA system via a two-branch Transformer structure. This structure, featuring a Twin-Transformer design, compactly combines 2D and 3D modalities and captures fine-grained correlations between modalities, allowing them mutually augmenting each other. Integrating proposed mechanisms above, we present BridgeQA, that offers a fresh perspective on multi-modal transformer-based architectures for 3D-VQA. Experiments validate that BridgeQA achieves state-of-the-art on 3D-VQA datasets and significantly outperforms existing solutions. Code is available at https://github.com/matthewdm0816/BridgeQA{this URL}.

Deep research agents have attracted increasing attention for their ability to collect large-scale online information to acquire target knowledge, with recent efforts shifting from purely text-based information seeking to multimodal settings. However, existing agentic workflows are largely aligned with evidence accumulation models, which linearly aggregate evidence and lack principled mechanisms for handling contradictory information across heterogeneous modalities. Towards this end, we propose Struct-Searcher, a structural agentic workflow grounded in belief revision theory that explicitly maintains an evolving multimodal structural graph throughout the reasoning process, enabling effective conflict-aware multimodal deep information seeking. Extensive experiments across multiple benchmark datasets and backbone models demonstrate that Struct-Searcher is (1) plug-and-play and model-agnostic, yielding an average relative accuracy improvement of 17.2% on BrowseComp-VL across five different backbones. (2) top-performing, consistently outperforming state-of-the-art vision-language models (VLMs) and deep research agents, with relative accuracy improvements of 3.7% on MM-BrowseComp, 1.5% on HLE-VL, and 0.7% on BrowseComp-VL over the second-best competing approach.

Monitoring complex assembly processes is critical for maintaining productivity and ensuring compliance with assembly standards. However, variability in human actions and subjective task preferences complicate accurate task anticipation and guidance. To address these challenges, we introduce the Multi-Modal Transformer Fusion and Recurrent Units (MMTFRU) Network for egocentric activity anticipation, utilizing multimodal fusion to improve prediction accuracy. Integrated with the Operator Action Monitoring Unit (OAMU), the system provides proactive operator guidance, preventing deviations in the assembly process. OAMU employs two strategies: (1) Top-5 MMTF-RU predictions, combined with a reference graph and an action dictionary, for next-step recommendations; and (2) Top-1 MMTF-RU predictions, integrated with a reference graph, for detecting sequence deviations and predicting anomaly scores via an entropy-informed confidence mechanism. We also introduce Time-Weighted Sequence Accuracy (TWSA) to evaluate operator efficiency and ensure timely task completion. Our approach is validated on the industrial Meccano dataset and the largescale EPIC-Kitchens-55 dataset, demonstrating its effectiveness in dynamic environments.

Clothes-Changing Re-Identification (CC-ReID) aims to recognize individuals across different locations and times, irrespective of clothing. Existing methods often rely on additional models or annotations to learn robust, clothing-invariant features, making them resource-intensive. In contrast, we explore the use of color - specifically foreground and background colors - as a lightweight, annotation-free proxy for mitigating appearance bias in ReID models. We propose Colors See, Colors Ignore (CSCI), an RGB-only method that leverages color information directly from raw images or video frames. CSCI efficiently captures color-related appearance bias ('Color See') while disentangling it from identity-relevant ReID features ('Color Ignore'). To achieve this, we introduce S2A self-attention, a novel self-attention to prevent information leak between color and identity cues within the feature space. Our analysis shows a strong correspondence between learned color embeddings and clothing attributes, validating color as an effective proxy when explicit clothing labels are unavailable. We demonstrate the effectiveness of CSCI on both image and video ReID with extensive experiments on four CC-ReID datasets. We improve the baseline by Top-1 2.9% on LTCC and 5.0% on PRCC for image-based ReID, and 1.0% on CCVID and 2.5% on MeVID for video-based ReID without relying on additional supervision. Our results highlight the potential of color as a cost-effective solution for addressing appearance bias in CC-ReID. Github: https://github.com/ppriyank/ICCV-CSCI-Person-ReID.

AI assistants produce vulnerable code in 45% of security-relevant scenarios, introducing flaws into production systems at scale. Yet existing secure coding datasets fall short. They lack incident grounding, don't provide the scale modern training requires, and miss the operational security context developers need for production deployments. We present SecureCode v2.0, a production-grade dataset of 1,215 security-focused coding examples that passed structural validation and expert security review. Every example ties to actual documented security incidents with CVE references, provides vulnerable and secure implementations, demonstrates concrete attacks, and includes defense-in-depth operational guidance. The dataset covers 11 vulnerability categories (complete OWASP Top 10:2025 plus AI/ML Security Threats) across 11 languages (Python, JavaScript, Java, Go, PHP, C#, TypeScript, Ruby, Rust, Kotlin, and YAML for infrastructure-as-code). Our quality assurance framework ensures complete incident grounding. Each example includes SIEM integration strategies, infrastructure hardening recommendations (Docker, AppArmor, WAF configurations), and testing approaches using language-appropriate frameworks. The dataset uses a 4-turn conversational structure mirroring actual developer-AI interactions, escalating from basic implementations to advanced security considerations and defense-in-depth guidance. Our contributions: (1) 1,215 rigorously validated examples split into 989 training, 122 validation, and 104 test sets, (2) an automated validation framework ensuring dataset consistency, (3) a 4-turn conversational structure capturing realistic security workflows, (4) comprehensive operational security guidance with SIEM integration strategies, (5) complete language-specific implementation fidelity, and (6) open-source release of data, validation tools, and benchmarking protocols.

Most 2D human pose estimation benchmarks are nearly saturated, with the exception of crowded scenes. We introduce PMPose, a top-down 2D pose estimator that incorporates the probabilistic formulation and the mask-conditioning. PMPose improves crowded pose estimation without sacrificing performance on standard scenes. Building on this, we present BBoxMaskPose v2 (BMPv2) integrating PMPose and an enhanced SAM-based mask refinement module. BMPv2 surpasses state-of-the-art by 1.5 average precision (AP) points on COCO and 6 AP points on OCHuman, becoming the first method to exceed 50 AP on OCHuman. We demonstrate that BMP's 2D prompting of 3D model improves 3D pose estimation in crowded scenes and that advances in 2D pose quality directly benefit 3D estimation. Results on the new OCHuman-Pose dataset show that multi-person performance is more affected by pose prediction accuracy than by detection. The code, models, and data are available on https://MiraPurkrabek.github.io/BBox-Mask-Pose/.

We propose EAR, a query Expansion And Reranking approach for improving passage retrieval, with the application to open-domain question answering. EAR first applies a query expansion model to generate a diverse set of queries, and then uses a query reranker to select the ones that could lead to better retrieval results. Motivated by the observation that the best query expansion often is not picked by greedy decoding, EAR trains its reranker to predict the rank orders of the gold passages when issuing the expanded queries to a given retriever. By connecting better the query expansion model and retriever, EAR significantly enhances a traditional sparse retrieval method, BM25. Empirically, EAR improves top-5/20 accuracy by 3-8 and 5-10 points in in-domain and out-of-domain settings, respectively, when compared to a vanilla query expansion model, GAR, and a dense retrieval model, DPR.

A wide array of sequence models are built on a framework modeled after Transformers, comprising alternating sequence mixer and channel mixer layers. This paper studies a unifying matrix mixer view of sequence mixers that can be conceptualized as a linear map on the input sequence. This framework encompasses a broad range of well-known sequence models, including the self-attention of Transformers as well as recent strong alternatives such as structured state space models (SSMs), and allows understanding downstream characteristics such as efficiency and expressivity through properties of their structured matrix class. We identify a key axis of matrix parameterizations termed sequence alignment, which increases the flexibility and performance of matrix mixers, providing insights into the strong performance of Transformers and recent SSMs such as Mamba. Furthermore, the matrix mixer framework offers a systematic approach to developing sequence mixers with desired properties, allowing us to develop several new sub-quadratic sequence models. In particular, we propose a natural bidirectional extension of the Mamba model (Hydra), parameterized as a quasiseparable matrix mixer, which demonstrates superior performance over other sequence models including Transformers on non-causal tasks. As a drop-in replacement for attention layers, Hydra outperforms BERT by 0.8 points on the GLUE benchmark and ViT by 2% Top-1 accuracy on ImageNet.

The field of Grammatical Error Correction (GEC) has produced various systems to deal with focused phenomena or general text editing. We propose an automatic way to combine black-box systems. Our method automatically detects the strength of a system or the combination of several systems per error type, improving precision and recall while optimizing F score directly. We show consistent improvement over the best standalone system in all the configurations tested. This approach also outperforms average ensembling of different RNN models with random initializations. In addition, we analyze the use of BERT for GEC - reporting promising results on this end. We also present a spellchecker created for this task which outperforms standard spellcheckers tested on the task of spellchecking. This paper describes a system submission to Building Educational Applications 2019 Shared Task: Grammatical Error Correction. Combining the output of top BEA 2019 shared task systems using our approach, currently holds the highest reported score in the open phase of the BEA 2019 shared task, improving F0.5 by 3.7 points over the best result reported.

Recent advances in knowledge distillation (KD) predominantly emphasize feature-level knowledge transfer, frequently overlooking critical information embedded within the teacher's logit distributions. In this paper, we revisit logit-based distillation and reveal an underexplored yet critical element: Top-K knowledge. Motivated by this insight, we propose Top-scaled Knowledge Distillation (TopKD), a simple, efficient, and architecture-agnostic framework that significantly enhances logit-based distillation. TopKD consists of two main components: (1) a Top-K Scaling Module (TSM), which adaptively amplifies the most informative logits, and (2) a Top-K Decoupled Loss (TDL), which offers targeted and effective supervision. Notably, TopKD integrates seamlessly into existing KD methods without introducing extra modules or requiring architectural changes. Extensive experiments on CIFAR-100, ImageNet, STL-10, and Tiny-ImageNet demonstrate that TopKD consistently surpasses state-of-the-art distillation methods. Moreover, our method demonstrates substantial effectiveness when distilling Vision Transformers, underscoring its versatility across diverse network architectures. These findings highlight the significant potential of logits to advance knowledge distillation.

AISHELL-1 is by far the largest open-source speech corpus available for Mandarin speech recognition research. It was released with a baseline system containing solid training and testing pipelines for Mandarin ASR. In AISHELL-2, 1000 hours of clean read-speech data from iOS is published, which is free for academic usage. On top of AISHELL-2 corpus, an improved recipe is developed and released, containing key components for industrial applications, such as Chinese word segmentation, flexible vocabulary expension and phone set transformation etc. Pipelines support various state-of-the-art techniques, such as time-delayed neural networks and Lattic-Free MMI objective funciton. In addition, we also release dev and test data from other channels(Android and Mic). For research community, we hope that AISHELL-2 corpus can be a solid resource for topics like transfer learning and robust ASR. For industry, we hope AISHELL-2 recipe can be a helpful reference for building meaningful industrial systems and products.

We present a CLSRIL-23, a self supervised learning based audio pre-trained model which learns cross lingual speech representations from raw audio across 23 Indic languages. It is built on top of wav2vec 2.0 which is solved by training a contrastive task over masked latent speech representations and jointly learns the quantization of latents shared across all languages. We compare the language wise loss during pretraining to compare effects of monolingual and multilingual pretraining. Performance on some downstream fine-tuning tasks for speech recognition is also compared and our experiments show that multilingual pretraining outperforms monolingual training, in terms of learning speech representations which encodes phonetic similarity of languages and also in terms of performance on down stream tasks. A decrease of 5% is observed in WER and 9.5% in CER when a multilingual pretrained model is used for finetuning in Hindi. All the code models are also open sourced. CLSRIL-23 is a model trained on 23 languages and almost 10,000 hours of audio data to facilitate research in speech recognition for Indic languages. We hope that new state of the art systems will be created using the self supervised approach, especially for low resources Indic languages.

MultiWOZ is a well-known task-oriented dialogue dataset containing over 10,000 annotated dialogues spanning 8 domains. It is extensively used as a benchmark for dialogue state tracking. However, recent works have reported presence of substantial noise in the dialogue state annotations. MultiWOZ 2.1 identified and fixed many of these erroneous annotations and user utterances, resulting in an improved version of this dataset. This work introduces MultiWOZ 2.2, which is a yet another improved version of this dataset. Firstly, we identify and fix dialogue state annotation errors across 17.3% of the utterances on top of MultiWOZ 2.1. Secondly, we redefine the ontology by disallowing vocabularies of slots with a large number of possible values (e.g., restaurant name, time of booking). In addition, we introduce slot span annotations for these slots to standardize them across recent models, which previously used custom string matching heuristics to generate them. We also benchmark a few state of the art dialogue state tracking models on the corrected dataset to facilitate comparison for future work. In the end, we discuss best practices for dialogue data collection that can help avoid annotation errors.

The ARC-AGI benchmark series serves as a critical measure of few-shot generalization on novel tasks, a core aspect of intelligence. The ARC Prize 2025 global competition targeted the newly released ARC-AGI-2 dataset, which features greater task complexity compared to its predecessor. The Kaggle competition attracted 1,455 teams and 15,154 entries, with the top score reaching 24% on the ARC-AGI-2 private evaluation set. Paper submissions nearly doubled year-over-year to 90 entries, reflecting the growing research interest in fluid intelligence and abstract reasoning. The defining theme of 2025 is the emergence of the refinement loop -- a per-task iterative program optimization loop guided by a feedback signal. Refinement loops come in a variety of forms, in particular evolutionary program synthesis approaches and application-layer refinements to commercial AI systems. Such refinement loops are also possible in weight space, as evidenced by zero-pretraining deep learning methods which are now achieving competitive performance with remarkably small networks (7M parameters). In parallel, four frontier AI labs (Anthropic, Google DeepMind, OpenAI, and xAI) reported ARC-AGI performance in public model cards in 2025, establishing ARC-AGI as an industry standard benchmark for AI reasoning. However, our analysis indicates that current frontier AI reasoning performance remains fundamentally constrained to knowledge coverage, giving rise to new forms of benchmark contamination. In this paper, we survey the top-performing methods, examine the role of refinement loops in AGI progress, discuss knowledge-dependent overfitting, and preview ARC-AGI-3, which introduces interactive reasoning challenges that require exploration, planning, memory, goal acquisition, and alignment capabilities.

Advancements in space telescopes have opened new avenues for gathering vast amounts of data on exoplanet atmosphere spectra. However, accurately extracting chemical and physical properties from these spectra poses significant challenges due to the non-linear nature of the underlying physics. This paper presents novel machine learning models developed by the AstroAI team for the Ariel Data Challenge 2023, where one of the models secured the top position among 293 competitors. Leveraging Normalizing Flows, our models predict the posterior probability distribution of atmospheric parameters under different atmospheric assumptions. Moreover, we introduce an alternative model that exhibits higher performance potential than the winning model, despite scoring lower in the challenge. These findings highlight the need to reevaluate the evaluation metric and prompt further exploration of more efficient and accurate approaches for exoplanet atmosphere spectra analysis. Finally, we present recommendations to enhance the challenge and models, providing valuable insights for future applications on real observational data. These advancements pave the way for more effective and timely analysis of exoplanet atmospheric properties, advancing our understanding of these distant worlds.

Optimization of Top-1 ImageNet promotes enormous networks that may be impractical in inference settings. Binary neural networks (BNNs) have the potential to significantly lower the compute intensity but existing models suffer from low quality. To overcome this deficiency, we propose PokeConv, a binary convolution block which improves quality of BNNs by techniques such as adding multiple residual paths, and tuning the activation function. We apply it to ResNet-50 and optimize ResNet's initial convolutional layer which is hard to binarize. We name the resulting network family PokeBNN. These techniques are chosen to yield favorable improvements in both top-1 accuracy and the network's cost. In order to enable joint optimization of the cost together with accuracy, we define arithmetic computation effort (ACE), a hardware- and energy-inspired cost metric for quantized and binarized networks. We also identify a need to optimize an under-explored hyper-parameter controlling the binarization gradient approximation. We establish a new, strong state-of-the-art (SOTA) on top-1 accuracy together with commonly-used CPU64 cost, ACE cost and network size metrics. ReActNet-Adam, the previous SOTA in BNNs, achieved a 70.5% top-1 accuracy with 7.9 ACE. A small variant of PokeBNN achieves 70.5% top-1 with 2.6 ACE, more than 3x reduction in cost; a larger PokeBNN achieves 75.6% top-1 with 7.8 ACE, more than 5% improvement in accuracy without increasing the cost. PokeBNN implementation in JAX/Flax and reproduction instructions are available in AQT repository: https://github.com/google/aqt

Modern action recognition models operate on memory- and compute-intensive dense RGB video volumes and frequently exploit appearance and background shortcuts, for example, predicting actions from objects or scenes instead of characteristic motion. We investigate an efficient alternative input modality that is largely free of such biases by construction: sparse point trajectories. To this end, we develop a simple transformer architecture for 2.5D trajectory-based recognition together with a masked-trajectory pretraining, which we show to substantially improve downstream action recognition accuracy. Despite using only a fraction of the dense RGB input, our method reaches 45% top-1 on Something-Something V2 and 54% on EPIC-Kitchens-100, and surpasses V-JEPA on time-reversal sensitivity. More importantly, we find trajectory features to be complementary to state-of-the-art appearance-based features. Fusing our pretrained model with DINOv2 and V-JEPA 2 improves top-1 accuracy on Something-Something V2 by 8.7 and 1.6 points, respectively. Code: https://github.com/ecker-lab/TrAction

As Large Language Models (LLMs) are increasingly deployed as autonomous agents in complex and long horizon settings, it is critical to evaluate their ability to sabotage users by pursuing hidden objectives. We study the ability of frontier LLMs to evade monitoring and achieve harmful hidden goals while completing a wide array of realistic tasks. We evaluate a broad range of frontier LLMs using SHADE (Subtle Harmful Agent Detection & Evaluation)-Arena, the first highly diverse agent evaluation dataset for sabotage and monitoring capabilities of LLM agents. SHADE-Arena consists of complex pairs of benign main tasks and harmful side objectives in complicated environments. Agents are evaluated on their ability to complete the side task without appearing suspicious to an LLM monitor. When measuring agent ability to (a) complete the main task, (b) complete the side task, and (c) avoid detection, we find that the best performing frontier models score 27% (Claude 3.7 Sonnet) and 15% (Gemini 2.5 Pro) as sabotage agents when overseen by Claude 3.6 Sonnet. For current frontier models, success on the side task relies heavily on having access to a hidden scratchpad that is not visible to the monitor. We also use SHADE-Arena to measure models' monitoring abilities, with the top monitor (Gemini 2.5 Pro) achieving an AUC of 0.87 at distinguishing benign and malign transcripts. We find that for now, models still struggle at sabotage due to failures in long-context main task execution. However, our measurements already demonstrate the difficulty of monitoring for subtle sabotage attempts, which we expect to only increase in the face of more complex and longer-horizon tasks.

Scientific artifacts such as models and datasets are foundations for research. With the rapid growth of platforms like HuggingFace, researchers now have access to a large number of artifacts. Yet, a key challenge remains: how can we automatically discover the state-of-the-art (SOTA) model for a given dataset by fully leveraging existing artifacts? We formalize this task as automatic SOTA discovery by modeling HuggingFace as an artifact graph, where nodes are models/datasets and edges represent evaluations. We propose ArtifactLinker, a two-stage framework: (1) ranking promising unobserved model--dataset links using Graph Neural Networks (GNNs) or graph-augmented Large Language Models (LLMs), and (2) verifying top-ranked links via coding experiments with LLM-based agents. We further introduce a benchmark named ArtifactBench with 14,053 artifacts and 51,337 relations to evaluate the performance of both stages. Results show that (1) graph structures between existing artifacts are effective for missing link prediction; (2) end-to-end ranking and verification with ArtifactLinker help discover potential SOTA results and research insights.

Large Language Models (LLMs) have gained widespread use in various applications due to their powerful capability to generate human-like text. However, prompt injection attacks, which involve overwriting a model's original instructions with malicious prompts to manipulate the generated text, have raised significant concerns about the security and reliability of LLMs. Ensuring that LLMs are robust against such attacks is crucial for their deployment in real-world applications, particularly in critical tasks. In this paper, we propose PROMPTFUZZ, a novel testing framework that leverages fuzzing techniques to systematically assess the robustness of LLMs against prompt injection attacks. Inspired by software fuzzing, PROMPTFUZZ selects promising seed prompts and generates a diverse set of prompt injections to evaluate the target LLM's resilience. PROMPTFUZZ operates in two stages: the prepare phase, which involves selecting promising initial seeds and collecting few-shot examples, and the focus phase, which uses the collected examples to generate diverse, high-quality prompt injections. Using PROMPTFUZZ, we can uncover more vulnerabilities in LLMs, even those with strong defense prompts. By deploying the generated attack prompts from PROMPTFUZZ in a real-world competition, we achieved the 7th ranking out of over 4000 participants (top 0.14%) within 2 hours. Additionally, we construct a dataset to fine-tune LLMs for enhanced robustness against prompt injection attacks. While the fine-tuned model shows improved robustness, PROMPTFUZZ continues to identify vulnerabilities, highlighting the importance of robust testing for LLMs. Our work emphasizes the critical need for effective testing tools and provides a practical framework for evaluating and improving the robustness of LLMs against prompt injection attacks.

Contextual cues related to a person's pose and interactions with objects and other people in the scene can provide valuable information for gaze following. While existing methods have focused on dedicated cue extraction methods, in this work we investigate the zero-shot capabilities of Vision-Language Models (VLMs) for extracting a wide array of contextual cues to improve gaze following performance. We first evaluate various VLMs, prompting strategies, and in-context learning (ICL) techniques for zero-shot cue recognition performance. We then use these insights to extract contextual cues for gaze following, and investigate their impact when incorporated into a state of the art model for the task. Our analysis indicates that BLIP-2 is the overall top performing VLM and that ICL can improve performance. We also observe that VLMs are sensitive to the choice of the text prompt although ensembling over multiple text prompts can provide more robust performance. Additionally, we discover that using the entire image along with an ellipse drawn around the target person is the most effective strategy for visual prompting. For gaze following, incorporating the extracted cues results in better generalization performance, especially when considering a larger set of cues, highlighting the potential of this approach.

Air pollution is a crucial issue affecting human health and livelihoods, as well as one of the barriers to economic and social growth. Forecasting air quality has become an increasingly important endeavor with significant social impacts, especially in emerging countries like China. In this paper, we present a novel Transformer architecture termed AirFormer to collectively predict nationwide air quality in China, with an unprecedented fine spatial granularity covering thousands of locations. AirFormer decouples the learning process into two stages -- 1) a bottom-up deterministic stage that contains two new types of self-attention mechanisms to efficiently learn spatio-temporal representations; 2) a top-down stochastic stage with latent variables to capture the intrinsic uncertainty of air quality data. We evaluate AirFormer with 4-year data from 1,085 stations in the Chinese Mainland. Compared to the state-of-the-art model, AirFormer reduces prediction errors by 5%~8% on 72-hour future predictions. Our source code is available at https://github.com/yoshall/airformer.

We show that cascaded diffusion models are capable of generating high fidelity images on the class-conditional ImageNet generation benchmark, without any assistance from auxiliary image classifiers to boost sample quality. A cascaded diffusion model comprises a pipeline of multiple diffusion models that generate images of increasing resolution, beginning with a standard diffusion model at the lowest resolution, followed by one or more super-resolution diffusion models that successively upsample the image and add higher resolution details. We find that the sample quality of a cascading pipeline relies crucially on conditioning augmentation, our proposed method of data augmentation of the lower resolution conditioning inputs to the super-resolution models. Our experiments show that conditioning augmentation prevents compounding error during sampling in a cascaded model, helping us to train cascading pipelines achieving FID scores of 1.48 at 64x64, 3.52 at 128x128 and 4.88 at 256x256 resolutions, outperforming BigGAN-deep, and classification accuracy scores of 63.02% (top-1) and 84.06% (top-5) at 256x256, outperforming VQ-VAE-2.

EarthVision Embed2Scale challenge (CVPR 2025) aims to develop foundational geospatial models to embed SSL4EO-S12 hyperspectral geospatial data cubes into embedding vectors that faciliatetes various downstream tasks, e.g., classification, regression, etc. In this technical report, we introduce our proposed method for the Top-1 winning solution on the Embed2Scale Challenge.

Human pose estimation (HPE) in the top-view using fisheye cameras presents a promising and innovative application domain. However, the availability of datasets capturing this viewpoint is extremely limited, especially those with high-quality 2D and 3D keypoint annotations. Addressing this gap, we leverage the capabilities of Neural Radiance Fields (NeRF) technique to establish a comprehensive pipeline for generating human pose datasets from existing 2D and 3D datasets, specifically tailored for the top-view fisheye perspective. Through this pipeline, we create a novel dataset NToP570K (NeRF-powered Top-view human Pose dataset for fisheye cameras with over 570 thousand images), and conduct an extensive evaluation of its efficacy in enhancing neural networks for 2D and 3D top-view human pose estimation. A pretrained ViTPose-B model achieves an improvement in AP of 33.3 % on our validation set for 2D HPE after finetuning on our training set. A similarly finetuned HybrIK-Transformer model gains 53.7 mm reduction in PA-MPJPE for 3D HPE on the validation set.

Large language models (LLMs) typically employ greedy decoding or low-temperature sampling for reasoning tasks, reflecting a perceived trade-off between diversity and accuracy. We challenge this convention by introducing top-nsigma, a novel sampling method that operates directly on pre-softmax logits by leveraging a statistical threshold. Our key insight is that logits naturally separate into a Gaussian-distributed noisy region and a distinct informative region, enabling efficient token filtering without complex probability manipulations. Unlike existing methods (e.g., top-p, min-p) that inadvertently include more noise tokens at higher temperatures, top-nsigma maintains a stable sampling space regardless of temperature scaling. We also provide a theoretical analysis of top-nsigma to better understand its behavior. The extensive experimental results across four reasoning-focused datasets demonstrate that our method not only outperforms existing sampling approaches but also surpasses greedy decoding, while maintaining consistent performance even at high temperatures.

Visual instruction tuning aims to enable large language models to comprehend the visual world, with a pivotal challenge lying in establishing an effective vision-to-language projection. However, existing methods often grapple with the intractable trade-off between accuracy and efficiency. In this paper, we present LLaVA-Meteor, a novel approach designed to break this deadlock, equipped with a novel Top-Down Compression paradigm that strategically compresses visual tokens without compromising core information. Specifically, we construct a trainable Flash Global Fusion module based on efficient selective state space operators, which aligns the feature space while enabling each token to perceive holistic visual context and instruction preference at low cost. Furthermore, a local-to-single scanning manner is employed to effectively capture local dependencies, thereby enhancing the model's capability in vision modeling. To alleviate computational overhead, we explore a Visual-Native Selection mechanism that independently assesses token significance by both the visual and native experts, followed by aggregation to retain the most critical subset. Extensive experiments show that our approach reduces visual tokens by 75--95% while achieving comparable or superior performance across 12 benchmarks, significantly improving efficiency.

We present Top-Theta (Top-theta) Attention, a training-free method for sparsifying transformer attention during inference. Our key insight is that static, per-head thresholds can be calibrated to retain the desired constant number of significant elements per attention row. This approach enables content-based sparsity without retraining, and it remains robust across data domains. We further introduce compensation techniques to preserve accuracy under aggressive sparsification, establishing attention thresholding as a practical and principled alternative to top-k attention. We provide extensive evaluation on natural language processing tasks, showing that Top-theta achieves 3-10x reduction in V-cache usage and up to 10x fewer attention elements during inference while degrading no more than 1% in accuracy.

LLMs have become the go-to choice for code generation tasks, with an exponential increase in the training, development, and usage of LLMs specifically for code generation. To evaluate the ability of LLMs on code, both academic and industry practitioners rely on popular handcrafted benchmarks. However, prior benchmarks contain only a very limited set of problems, both in quantity and variety. Further, due to popularity and age, many benchmarks are prone to data leakage where example solutions can be readily found on the web and thus potentially in training data. Such limitations inevitably lead us to inquire: Is the leaderboard performance on existing benchmarks reliable and comprehensive enough to measure the program synthesis ability of LLMs? To address this, we introduce EvoEval -- a program synthesis benchmark suite created by evolving existing benchmarks into different targeted domains for a comprehensive evaluation of LLM coding abilities. Our study on 51 LLMs shows that compared to the high performance obtained on standard benchmarks like HumanEval, there is a significant drop in performance (on average 39.4%) when using EvoEval. Additionally, the decrease in performance can range from 19.6% to 47.7%, leading to drastic ranking changes amongst LLMs and showing potential overfitting of existing benchmarks. Furthermore, we showcase various insights, including the brittleness of instruction-following models when encountering rewording or subtle changes as well as the importance of learning problem composition and decomposition. EvoEval not only provides comprehensive benchmarks, but can be used to further evolve arbitrary problems to keep up with advances and the ever-changing landscape of LLMs for code. We have open-sourced our benchmarks, tools, and complete LLM generations at https://github.com/evo-eval/evoeval

Top-K sparse softmax gating mixture of experts has been widely used for scaling up massive deep-learning architectures without increasing the computational cost. Despite its popularity in real-world applications, the theoretical understanding of that gating function has remained an open problem. The main challenge comes from the structure of the top-K sparse softmax gating function, which partitions the input space into multiple regions with distinct behaviors. By focusing on a Gaussian mixture of experts, we establish theoretical results on the effects of the top-K sparse softmax gating function on both density and parameter estimations. Our results hinge upon defining novel loss functions among parameters to capture different behaviors of the input regions. When the true number of experts k_{ast} is known, we demonstrate that the convergence rates of density and parameter estimations are both parametric on the sample size. However, when k_{ast} becomes unknown and the true model is over-specified by a Gaussian mixture of k experts where k > k_{ast}, our findings suggest that the number of experts selected from the top-K sparse softmax gating function must exceed the total cardinality of a certain number of Voronoi cells associated with the true parameters to guarantee the convergence of the density estimation. Moreover, while the density estimation rate remains parametric under this setting, the parameter estimation rates become substantially slow due to an intrinsic interaction between the softmax gating and expert functions.

Large language models (LLMs), despite their impressive performance across a wide range of tasks, often struggle to balance two competing objectives in open-ended text generation: fostering diversity and creativity while preserving logical coherence. Existing truncated sampling techniques, including temperature scaling, top-\p (nucleus) sampling, and min-\p sampling, aim to manage this trade-off. However, they exhibit limitations, particularly in the effective incorporation of the confidence of the model into the corresponding sampling strategy. For example, min-\p sampling relies on a single top token as a heuristic for confidence, eventually underutilizing the information of the probability distribution. Toward effective incorporation of the confidence of the model, in this paper, we present **top-H** decoding. We first establish the theoretical foundation of the interplay between creativity and coherence in truncated sampling by formulating an **entropy-constrained minimum divergence** problem. We then prove this minimization problem to be equivalent to an **entropy-constrained mass maximization** (ECMM) problem, which is NP-hard. Finally, we present top-H decoding, a computationally efficient greedy algorithm to solve the ECMM problem. Extensive empirical evaluations demonstrate that top-H outperforms the state-of-the-art (SoTA) alternative of min-\p sampling by up to **25.63%** on creative writing benchmarks, while maintaining robustness on question-answering datasets such as GPQA, GSM8K, and MT-Bench. Additionally, an *LLM-as-judge* evaluation confirms that top-H indeed produces coherent outputs even at higher temperatures, where creativity is especially critical. In summary, top-H advances SoTA in open-ended text generation and can be *easily integrated* into creative writing applications. The code is available at https://github.com/ErfanBaghaei/Top-H-Decoding.

Top-k decoding is a widely used method for sampling from LLMs: at each token, only the largest k next-token-probabilities are kept, and the next token is sampled after re-normalizing them to sum to unity. Top-k and other sampling methods are motivated by the intuition that true next-token distributions are sparse, and the noisy LLM probabilities need to be truncated. However, to our knowledge, a precise theoretical motivation for the use of top-k decoding is missing. In this work, we develop a theoretical framework that both explains and generalizes top-k decoding. We view decoding at a fixed token as the recovery of a sparse probability distribution. We consider Bregman decoders obtained by minimizing a separable Bregman divergence (for both the primal and dual cases) with a sparsity-inducing ell_0 regularization. Despite the combinatorial nature of the objective, we show how to optimize it efficiently for a large class of divergences. We show that the optimal decoding strategies are greedy, and further that the loss function is discretely convex in k, so that binary search provably and efficiently finds the optimal k. We show that top-k decoding arises as a special case for the KL divergence, and identify new decoding strategies that have distinct behaviors (e.g., non-linearly up-weighting larger probabilities after re-normalization).

Top-view perspective denotes a typical way in which humans read and reason over different types of maps, and it is vital for localization and navigation of humans as well as of `non-human' agents, such as the ones backed by large Vision-Language Models (VLMs). Nonetheless, spatial reasoning capabilities of modern VLMs remain unattested and underexplored. In this work, we thus study their capability to understand and reason over spatial relations from the top view. The focus on top view also enables controlled evaluations at different granularity of spatial reasoning; we clearly disentangle different abilities (e.g., recognizing particular objects versus understanding their relative positions). We introduce the TopViewRS (Top-View Reasoning in Space) dataset, consisting of 11,384 multiple-choice questions with either realistic or semantic top-view map as visual input. We then use it to study and evaluate VLMs across 4 perception and reasoning tasks with different levels of complexity. Evaluation of 10 representative open- and closed-source VLMs reveals the gap of more than 50% compared to average human performance, and it is even lower than the random baseline in some cases. Although additional experiments show that Chain-of-Thought reasoning can boost model capabilities by 5.82% on average, the overall performance of VLMs remains limited. Our findings underscore the critical need for enhanced model capability in top-view spatial reasoning and set a foundation for further research towards human-level proficiency of VLMs in real-world multimodal tasks.

We study extractive question-answering in the medical domain (Medical-EQA). This problem has two main challenges: (i) domain specificity, as most AI models lack necessary domain knowledge, and (ii) extraction-based answering style, which restricts most autoregressive LLMs due to potential hallucinations. To handle those challenges, we propose TOP-Training, a target-oriented pre-training paradigm that stands out among all domain adaptation techniques with two desirable features: (i) TOP-Training moves one step further than popular domain-oriented fine-tuning since it not only moves closer to the target domain, but also familiarizes itself with the target dataset, and (ii) it does not assume the existence of a large set of unlabeled instances from the target domain. Specifically, for a target Medical-EQA dataset, we extract its entities and leverage large language models (LLMs) to generate synthetic texts containing those entities; we then demonstrate that pretraining on this synthetic text data yields better performance on the target Medical-EQA benchmarks. Overall, our contributions are threefold: (i) TOP-Training, a new pretraining technique to effectively adapt LLMs to better solve a target problem, (ii) TOP-Training has a wide application scope because it does not require the target problem to have a large set of unlabeled data, and (iii) our experiments highlight the limitations of autoregressive LLMs, emphasizing TOP-Training as a means to unlock the true potential of bidirectional LLMs.

We present compact analytic results for tree-level amplitudes containing a t t pair accompanied by up to four massless partons, t tgg, t tggg, t tgggg, t tqq, t tqqg, t tqqgg and ttqqq^prime q^prime. The results, obtained using BCFW on-shell recursion, are based both on previous published results and on the new calculations performed in this paper. These amplitudes are sufficient to calculate the production of a tt pair and zero, one, or two light parton jets, with the option to include the tree-level decays t to b νe^+ and t to b e^- barν efficiently. Our results are part of the NNLO corrections to t t production including the decay correlations for on-shell top quarks.

The top-down and bottom-up methods are two mainstreams of referring segmentation, while both methods have their own intrinsic weaknesses. Top-down methods are chiefly disturbed by Polar Negative (PN) errors owing to the lack of fine-grained cross-modal alignment. Bottom-up methods are mainly perturbed by Inferior Positive (IP) errors due to the lack of prior object information. Nevertheless, we discover that two types of methods are highly complementary for restraining respective weaknesses but the direct average combination leads to harmful interference. In this context, we build Win-win Cooperation (WiCo) to exploit complementary nature of two types of methods on both interaction and integration aspects for achieving a win-win improvement. For the interaction aspect, Complementary Feature Interaction (CFI) provides fine-grained information to top-down branch and introduces prior object information to bottom-up branch for complementary feature enhancement. For the integration aspect, Gaussian Scoring Integration (GSI) models the gaussian performance distributions of two branches and weightedly integrates results by sampling confident scores from the distributions. With our WiCo, several prominent top-down and bottom-up combinations achieve remarkable improvements on three common datasets with reasonable extra costs, which justifies effectiveness and generality of our method.

Following the success of dot-product attention in Transformers, numerous approximations have been recently proposed to address its quadratic complexity with respect to the input length. While these variants are memory and compute efficient, it is not possible to directly use them with popular pre-trained language models trained using vanilla attention, without an expensive corrective pre-training stage. In this work, we propose a simple yet highly accurate approximation for vanilla attention. We process the queries in chunks, and for each query, compute the top-k scores with respect to the keys. Our approach offers several advantages: (a) its memory usage is linear in the input size, similar to linear attention variants, such as Performer and RFA (b) it is a drop-in replacement for vanilla attention that does not require any corrective pre-training, and (c) it can also lead to significant memory savings in the feed-forward layers after casting them into the familiar query-key-value framework. We evaluate the quality of top-k approximation for multi-head attention layers on the Long Range Arena Benchmark, and for feed-forward layers of T5 and UnifiedQA on multiple QA datasets. We show our approach leads to accuracy that is nearly-identical to vanilla attention in multiple setups including training from scratch, fine-tuning, and zero-shot inference.

As long-context inference becomes central to large language models (LLMs), attention over growing key-value caches emerges as a dominant decoding bottleneck, motivating sparse attention for scalable inference. Fixed-budget top-k sparse attention cannot adapt to heterogeneous attention distributions across heads and layers, whereas top-p sparse attention directly preserves attention mass and provides stronger accuracy guarantees. Existing top-p methods, however, fail to jointly optimize top-p accuracy, selection overhead, and sparse attention cost, which limits their overall efficiency. We present Double-P, a hierarchical sparse attention framework that optimizes all three stages. Double-P first performs coarse-grained top-p estimation at the cluster level using size-weighted centroids, then adaptively refines computation through a second top-p stage that allocates token-level attention only when needed. Across long-context benchmarks, Double-P consistently achieves near-zero accuracy drop, reducing attention computation overhead by up to 1.8x and delivers up to 1.3x end-to-end decoding speedup over state-of-the-art fixed-budget sparse attention methods.

Most models of visual attention aim at predicting either top-down or bottom-up control, as studied using different visual search and free-viewing tasks. In this paper we propose the Human Attention Transformer (HAT), a single model that predicts both forms of attention control. HAT uses a novel transformer-based architecture and a simplified foveated retina that collectively create a spatio-temporal awareness akin to the dynamic visual working memory of humans. HAT not only establishes a new state-of-the-art in predicting the scanpath of fixations made during target-present and target-absent visual search and ``taskless'' free viewing, but also makes human gaze behavior interpretable. Unlike previous methods that rely on a coarse grid of fixation cells and experience information loss due to fixation discretization, HAT features a sequential dense prediction architecture and outputs a dense heatmap for each fixation, thus avoiding discretizing fixations. HAT sets a new standard in computational attention, which emphasizes effectiveness, generality, and interpretability. HAT's demonstrated scope and applicability will likely inspire the development of new attention models that can better predict human behavior in various attention-demanding scenarios. Code is available at https://github.com/cvlab-stonybrook/HAT.

Image Quality Assessment (IQA) is a fundamental task in computer vision that has witnessed remarkable progress with deep neural networks. Inspired by the characteristics of the human visual system, existing methods typically use a combination of global and local representations (\ie, multi-scale features) to achieve superior performance. However, most of them adopt simple linear fusion of multi-scale features, and neglect their possibly complex relationship and interaction. In contrast, humans typically first form a global impression to locate important regions and then focus on local details in those regions. We therefore propose a top-down approach that uses high-level semantics to guide the IQA network to focus on semantically important local distortion regions, named as TOPIQ. Our approach to IQA involves the design of a heuristic coarse-to-fine network (CFANet) that leverages multi-scale features and progressively propagates multi-level semantic information to low-level representations in a top-down manner. A key component of our approach is the proposed cross-scale attention mechanism, which calculates attention maps for lower level features guided by higher level features. This mechanism emphasizes active semantic regions for low-level distortions, thereby improving performance. CFANet can be used for both Full-Reference (FR) and No-Reference (NR) IQA. We use ResNet50 as its backbone and demonstrate that CFANet achieves better or competitive performance on most public FR and NR benchmarks compared with state-of-the-art methods based on vision transformers, while being much more efficient (with only {sim}13% FLOPS of the current best FR method). Codes are released at https://github.com/chaofengc/IQA-PyTorch.

Crowdsourcing has emerged as an effective platform for labeling large amounts of data in a cost- and time-efficient manner. Most previous work has focused on designing an efficient algorithm to recover only the ground-truth labels of the data. In this paper, we consider multi-choice crowdsourcing tasks with the goal of recovering not only the ground truth, but also the most confusing answer and the confusion probability. The most confusing answer provides useful information about the task by revealing the most plausible answer other than the ground truth and how plausible it is. To theoretically analyze such scenarios, we propose a model in which there are the top two plausible answers for each task, distinguished from the rest of the choices. Task difficulty is quantified by the probability of confusion between the top two, and worker reliability is quantified by the probability of giving an answer among the top two. Under this model, we propose a two-stage inference algorithm to infer both the top two answers and the confusion probability. We show that our algorithm achieves the minimax optimal convergence rate. We conduct both synthetic and real data experiments and demonstrate that our algorithm outperforms other recent algorithms. We also show the applicability of our algorithms in inferring the difficulty of tasks and in training neural networks with top-two soft labels.

Benefitting from insensitivity to light and high penetration of foggy environments, infrared cameras are widely used for sensing in nighttime traffic scenes. However, the low contrast and lack of chromaticity of thermal infrared (TIR) images hinder the human interpretation and portability of high-level computer vision algorithms. Colorization to translate a nighttime TIR image into a daytime color (NTIR2DC) image may be a promising way to facilitate nighttime scene perception. Despite recent impressive advances in image translation, semantic encoding entanglement and geometric distortion in the NTIR2DC task remain under-addressed. Hence, we propose a toP-down attEntion And gRadient aLignment based GAN, referred to as PearlGAN. A top-down guided attention module and an elaborate attentional loss are first designed to reduce the semantic encoding ambiguity during translation. Then, a structured gradient alignment loss is introduced to encourage edge consistency between the translated and input images. In addition, pixel-level annotation is carried out on a subset of FLIR and KAIST datasets to evaluate the semantic preservation performance of multiple translation methods. Furthermore, a new metric is devised to evaluate the geometric consistency in the translation process. Extensive experiments demonstrate the superiority of the proposed PearlGAN over other image translation methods for the NTIR2DC task. The source code and labeled segmentation masks will be available at https://github.com/FuyaLuo/PearlGAN/.

In monocular video 3D multi-person pose estimation, inter-person occlusion and close interactions can cause human detection to be erroneous and human-joints grouping to be unreliable. Existing top-down methods rely on human detection and thus suffer from these problems. Existing bottom-up methods do not use human detection, but they process all persons at once at the same scale, causing them to be sensitive to multiple-persons scale variations. To address these challenges, we propose the integration of top-down and bottom-up approaches to exploit their strengths. Our top-down network estimates human joints from all persons instead of one in an image patch, making it robust to possible erroneous bounding boxes. Our bottom-up network incorporates human-detection based normalized heatmaps, allowing the network to be more robust in handling scale variations. Finally, the estimated 3D poses from the top-down and bottom-up networks are fed into our integration network for final 3D poses. Besides the integration of top-down and bottom-up networks, unlike existing pose discriminators that are designed solely for single person, and consequently cannot assess natural inter-person interactions, we propose a two-person pose discriminator that enforces natural two-person interactions. Lastly, we also apply a semi-supervised method to overcome the 3D ground-truth data scarcity. Our quantitative and qualitative evaluations show the effectiveness of our method compared to the state-of-the-art baselines.

Retrieval-Augmented Generation (RAG) systems have been popular for generative applications, powering language models by injecting external knowledge. Companies have been trying to leverage their large catalog of documents (e.g. PDFs, presentation slides) in such RAG pipelines, whose first step is the retrieval component. Dense retrieval has been a popular approach, where embedding models are used to generate a dense representation of the user query that is closer to relevant content embeddings. More recently, VLM-based embedding models have become popular for visual document retrieval, as they preserve visual information and simplify the indexing pipeline compared to OCR text extraction. Motivated by the growing demand for visual document retrieval, we introduce Nemotron ColEmbed V2, a family of models that achieve state-of-the-art performance on the ViDoRe benchmarks. We release three variants - with 3B, 4B, and 8B parameters - based on pre-trained VLMs: NVIDIA Eagle 2 with Llama 3.2 3B backbone, Qwen3-VL-4B-Instruct and Qwen3-VL-8B-Instruct, respectively. The 8B model ranks first on the ViDoRe V3 leaderboard as of February 03, 2026, achieving an average NDCG@10 of 63.42. We describe the main techniques used across data processing, training, and post-training - such as cluster-based sampling, hard-negative mining, bidirectional attention, late interaction, and model merging - that helped us build our top-performing models. We also discuss compute and storage engineering challenges posed by the late interaction mechanism and present experiments on how to balance accuracy and storage with lower dimension embeddings.

We develop a unified mathematical framework for certified Top-k attention truncation that quantifies approximation error at both the distribution and output levels. For a single attention distribution P and its Top-k truncation hat P, we show that the total-variation distance coincides with the discarded softmax tail mass and satisfies TV(P,hat P)=1-e^{-KL(hat PVert P)}, yielding sharp Top-k-specific bounds in place of generic inequalities. From this we derive non-asymptotic deterministic bounds -- from a single boundary gap through multi-gap and blockwise variants -- that control TV(P,hat P) using only the ordered logits. Using an exact head-tail decomposition, we prove that the output error factorizes as |Attn(q,K,V)-Attn_k(q,K,V)|_2=τ|μ_{tail}-μ_{head}|_2 with τ=TV(P,hat P), yielding a new head-tail diameter bound |Attn(q,K,V)-Attn_k(q,K,V)|_2leτ,diam_{H,T} and refinements linking the error to Var_P(V). Under an i.i.d. Gaussian score model s_isimmathcal N(μ,σ^2) we derive closed-form tail masses and an asymptotic rule for the minimal k_varepsilon ensuring TV(P,hat P)levarepsilon, namely k_varepsilon/napproxΦ_c(σ+Φ^{-1}(varepsilon)). Experiments on bert-base-uncased and synthetic logits confirm the predicted scaling of k_varepsilon/n and show that certified Top-k can reduce scored keys by 2-4times on average while meeting the prescribed total-variation budget.

When rows of an n times d matrix A are given in a stream, we study algorithms for approximating the top eigenvector of the matrix {A}^TA (equivalently, the top right singular vector of A). We consider worst case inputs A but assume that the rows are presented to the streaming algorithm in a uniformly random order. We show that when the gap parameter R = σ_1(A)^2/σ_2(A)^2 = Ω(1), then there is a randomized algorithm that uses O(h cdot d cdot polylog(d)) bits of space and outputs a unit vector v that has a correlation 1 - O(1/R) with the top eigenvector v_1. Here h denotes the number of heavy rows in the matrix, defined as the rows with Euclidean norm at least |{A}|_F/d cdot operatorname{polylog(d)}. We also provide a lower bound showing that any algorithm using O(hd/R) bits of space can obtain at most 1 - Ω(1/R^2) correlation with the top eigenvector. Thus, parameterizing the space complexity in terms of the number of heavy rows is necessary for high accuracy solutions. Our results improve upon the R = Ω(log n cdot log d) requirement in a recent work of Price and Xun (FOCS 2024). We note that the algorithm of Price and Xun works for arbitrary order streams whereas our algorithm requires a stronger assumption that the rows are presented in a uniformly random order. We additionally show that the gap requirements in their analysis can be brought down to R = Ω(log^2 d) for arbitrary order streams and R = Ω(log d) for random order streams. The requirement of R = Ω(log d) for random order streams is nearly tight for their analysis as we obtain a simple instance with R = Ω(log d/loglog d) for which their algorithm, with any fixed learning rate, cannot output a vector approximating the top eigenvector v_1.

Large language models (LLMs) for code have become indispensable in various domains, including code generation, reasoning tasks and agent systems.While open-access code LLMs are increasingly approaching the performance levels of proprietary models, high-quality code LLMs suitable for rigorous scientific investigation, particularly those with reproducible data processing pipelines and transparent training protocols, remain limited. The scarcity is due to various challenges, including resource constraints, ethical considerations, and the competitive advantages of keeping models advanced. To address the gap, we introduce OpenCoder, a top-tier code LLM that not only achieves performance comparable to leading models but also serves as an ``open cookbook'' for the research community. Unlike most prior efforts, we release not only model weights and inference code, but also the reproducible training data, complete data processing pipeline, rigorous experimental ablation results, and detailed training protocols for open scientific research. Through this comprehensive release, we identify the key ingredients for building a top-tier code LLM: (1) code optimized heuristic rules for data cleaning and methods for data deduplication, (2) recall of text corpus related to code and (3) high-quality synthetic data in both annealing and supervised fine-tuning stages. By offering this level of openness, we aim to broaden access to all aspects of a top-tier code LLM, with OpenCoder serving as both a powerful model and an open foundation to accelerate research, and enable reproducible advancements in code AI.

As the application of Large Language Models (LLMs) expands, the demand for reliable evaluations increases. Existing LLM evaluation benchmarks primarily rely on static datasets, making it challenging to assess model performance in dynamic interactions with users. Moreover, these benchmarks often depend on specific background knowledge, complicating the measurement of a model's logical reasoning capabilities. Other dynamic evaluation methods based on strong models or manual efforts may introduce biases and incur high costs and time demands, hindering large-scale application. To address these issues, we propose TurtleBench. TurtleBench collects real user guesses from our online Turtle Soup Puzzle platform that we developed. This approach allows for the relatively dynamic generation of evaluation datasets, mitigating the risk of model cheating while aligning assessments more closely with genuine user needs for reasoning capabilities, thus enhancing the reliability of evaluations. TurtleBench includes 1,532 user guesses along with the correctness of guesses after annotation. Using this dataset, we thoroughly evaluated nine of the most advanced LLMs available today. Notably, the OpenAI o1 series models did not achieve leading results in these evaluations. We propose several hypotheses for further research, such as "the latent reasoning of o1 utilizes trivial Chain-of-Thought (CoT) techniques" and "increasing CoT length not only provides reasoning benefits but also incurs noise costs."

In this paper, we identify and characterize the emerging area of representation engineering (RepE), an approach to enhancing the transparency of AI systems that draws on insights from cognitive neuroscience. RepE places population-level representations, rather than neurons or circuits, at the center of analysis, equipping us with novel methods for monitoring and manipulating high-level cognitive phenomena in deep neural networks (DNNs). We provide baselines and an initial analysis of RepE techniques, showing that they offer simple yet effective solutions for improving our understanding and control of large language models. We showcase how these methods can provide traction on a wide range of safety-relevant problems, including honesty, harmlessness, power-seeking, and more, demonstrating the promise of top-down transparency research. We hope that this work catalyzes further exploration of RepE and fosters advancements in the transparency and safety of AI systems.

In the realm of recommender systems (RS), Top-K ranking metrics such as NDCG@K are the gold standard for evaluating recommendation performance. However, during the training of recommendation models, optimizing NDCG@K poses significant challenges due to its inherent discontinuous nature and the intricate Top-K truncation. Recent efforts to optimize NDCG@K have either overlooked the Top-K truncation or suffered from high computational costs and training instability. To overcome these limitations, we propose SoftmaxLoss@K (SL@K), a novel recommendation loss tailored for NDCG@K optimization. Specifically, we integrate the quantile technique to handle Top-K truncation and derive a smooth upper bound for optimizing NDCG@K to address discontinuity. The resulting SL@K loss has several desirable properties, including theoretical guarantees, ease of implementation, computational efficiency, gradient stability, and noise robustness. Extensive experiments on four real-world datasets and three recommendation backbones demonstrate that SL@K outperforms existing losses with a notable average improvement of 6.03%. The code is available at https://github.com/Tiny-Snow/IR-Benchmark.

Motivated by the growing demand for retrieval systems that operate across modalities, we introduce llama-nemoretriever-colembed, a unified text-image retrieval model that delivers state-of-the-art performance across multiple benchmarks. We release two model variants, 1B and 3B. The 3B model achieves state of the art performance, scoring NDCG@5 91.0 on ViDoRe V1 and 63.5 on ViDoRe V2, placing first on both leaderboards as of June 27, 2025. Our approach leverages the NVIDIA Eagle2 Vision-Language model (VLM), modifies its architecture by replacing causal attention with bidirectional attention, and integrates a ColBERT-style late interaction mechanism to enable fine-grained multimodal retrieval in a shared embedding space. While this mechanism delivers superior retrieval accuracy, it introduces trade-offs in storage and efficiency. We provide a comprehensive analysis of these trade-offs. Additionally, we adopt a two-stage training strategy to enhance the model's retrieval capabilities.

Are Capture-the-Flag competitions obsolete? In 2025, Cybersecurity AI (CAI) systematically conquered some of the world's most prestigious hacking competitions, achieving Rank #1 at multiple events and consistently outperforming thousands of human teams. Across five major circuits-HTB's AI vs Humans, Cyber Apocalypse (8,129 teams), Dragos OT CTF, UWSP Pointer Overflow, and the Neurogrid CTF showdown-CAI demonstrated that Jeopardy-style CTFs have become a solved game for well-engineered AI agents. At Neurogrid, CAI captured 41/45 flags to claim the 50,000 top prize; at Dragos OT, it sprinted 37% faster to 10K points than elite human teams; even when deliberately paused mid-competition, it maintained top-tier rankings. Critically, CAI achieved this dominance through our specialized alias1 model architecture, which delivers enterprise-scale AI security operations at unprecedented cost efficiency and with augmented autonomy-reducing 1B token inference costs from 5,940 to just $119, making continuous security agent operation financially viable for the first time. These results force an uncomfortable reckoning: if autonomous agents now dominate competitions designed to identify top security talent at negligible cost, what are CTFs actually measuring? This paper presents comprehensive evidence of AI capability across the 2025 CTF circuit and argues that the security community must urgently transition from Jeopardy-style contests to Attack & Defense formats that genuinely test adaptive reasoning and resilience-capabilities that remain uniquely human, for now.

The production of a top-quark pair, the heaviest known elementary particle, in association with a light jet is a key process for studying the properties of the Standard Model of Particle Physics. Due to its significance as a signal process with considerable sensitivity to the top-quark mass and as a background process for new physics searches, it is crucial to predict differential cross sections with high precision. In this article, we present, for the first time, predictions for various kinematical observables at next-to-next-to-leading order in Quantum Chromodynamics. The perturbative behavior is analyzed, and uncertainties arising from missing higher-order contributions are substantially reduced. The necessary two-loop amplitudes have been evaluated in the leading-color approximation, and we provide estimates for the impact of the missing contributions.

Large Language Models (LLMs) have emerged as a pivotal research area, yet the attention module remains a critical bottleneck in LLM inference, even with techniques like KVCache to mitigate redundant computations. While various top-k attention mechanisms have been proposed to accelerate LLM inference by exploiting the inherent sparsity of attention, they often struggled to strike a balance between efficiency and accuracy. In this paper, we introduce HATA (Hash-Aware Top-k Attention), a novel approach that systematically integrates low-overhead learning-to-hash techniques into the Top-k attention process. Different from the existing top-k attention methods which are devoted to seeking an absolute estimation of qk score, typically with a great cost, HATA maps queries and keys into binary hash codes, and acquires the relative qk score order with a quite low cost, which is sufficient for realizing top-k attention. Extensive experiments demonstrate that HATA achieves up to 7.2times speedup compared to vanilla full attention while maintaining model accuracy. In addition, HATA outperforms the state-of-the-art top-k attention methods in both accuracy and efficiency across multiple mainstream LLM models and diverse tasks. HATA is open source at https://github.com/gpzlx1/HATA.

Most LLM-based agent frameworks adopt a top-down philosophy: humans decompose tasks, define workflows, and assign agents to execute each step. While effective on benchmark-style tasks, such systems rely on designer updates and overlook agents' potential to learn from experience. Recently, Silver and Sutton(2025) envision a shift into a new era, where agents could progress from a stream of experiences. In this paper, we instantiate this vision of experience-driven learning by introducing a bottom-up agent paradigm that mirrors the human learning process. Agents acquire competence through a trial-and-reasoning mechanism-exploring, reflecting on outcomes, and abstracting skills over time. Once acquired, skills can be rapidly shared and extended, enabling continual evolution rather than static replication. As more agents are deployed, their diverse experiences accelerate this collective process, making bottom-up design especially suited for open-ended environments. We evaluate this paradigm in Slay the Spire and Civilization V, where agents perceive through raw visual inputs and act via mouse outputs, the same as human players. Using a unified, game-agnostic codebase without any game-specific prompts or privileged APIs, our bottom-up agents acquire skills entirely through autonomous interaction, demonstrating the potential of the bottom-up paradigm in complex, real-world environments. Our code is available at https://github.com/AngusDujw/Bottom-Up-Agent.

The Transformer architecture has emerged as the dominant choice for building large language models (LLMs). However, with new LLMs emerging on a frequent basis, it is important to consider the potential value of architecture-agnostic approaches that can provide interpretability across a variety of architectures. Despite recent successes in the interpretability of LLMs, many existing approaches rely on complex methods that are often tied to a specific model design and come with a significant computational cost. To address these limitations, we propose a novel technique, called NormXLogit, for assessing the significance of individual input tokens. This method operates based on the input and output representations associated with each token. First, we demonstrate that during the pre-training of LLMs, the norms of word embeddings capture the importance of input tokens. Second, we reveal a significant relationship between a token's importance and the extent to which its representation can resemble the model's final prediction. Through extensive analysis, we show that our approach consistently outperforms existing gradient-based methods in terms of faithfulness. Additionally, our method achieves better performance in layer-wise explanations compared to the most prominent architecture-specific methods.

While existing machine learning (ML) frameworks focus on established platforms, like running CUDA on server-grade GPUs, there have been growing demands to enable emerging AI applications in a broader set of scenarios, such as running Large Language Models (LLMs) within browsers and mobile phones. However, deploying emerging models on new platforms (such as Metal and WebGPU) presents significant software engineering challenges due to rapid model evolution and limited tooling and practices for these platforms. Previous practice for ML model deployment often follows a bottom-up fashion, where engineers first implement individual required operators and then put them together. However, this traditional development approach fails to meet the productivity requirements when deploying emerging ML applications, with the testing and debugging part as a bottleneck. To this end, we introduce TapML, a top-down approach designed to streamline model deployment on diverse platforms. While the traditional bottom-up approach requires crafting manual tests, TapML automatically creates high-quality, realistic test data through operator-wise test carving. Furthermore, TapML uses a migration-based strategy to gradually offload model implementation from the mature source platform to the target platform, minimizing the debugging scope of compound errors. TapML has been used as the default development method in the MLC-LLM project to deploy emerging ML models. Within 2 years, TapML has accelerated the deployment of 105 emerging models in 27 model architectures across 5 emerging platforms. We show that TapML effectively boosts developer productivity while ensuring the quality of deployed models. Furthermore, we summarize comprehensive case studies from our real-world development, offering best practices for developing emerging ML systems.

This paper reveals the phenomenon of parameter heterogeneity in large language models (LLMs). We find that a small subset of ``cherry'' parameters exhibit a disproportionately large influence on model performance, while the vast majority of parameters have minimal impact. This heterogeneity is found to be prevalent across different model families, scales, and types. Motivated by this observation, we propose CherryQ, a novel quantization method that unifies the optimization of mixed-precision parameters. CherryQ identifies and preserves the critical cherry parameters in high precision while aggressively quantizing the remaining parameters to low precision. Extensive experiments demonstrate the effectiveness of CherryQ. CherryQ outperforms existing quantization approaches in terms of perplexity and downstream task performance. Notably, our 3-bit quantized Vicuna-1.5 exhibits competitive performance compared to their 16-bit counterparts. These findings highlight the potential of CherryQ for enabling efficient deployment of LLMs by taking advantage of parameter heterogeneity.

Hierarchical clustering of networks consists in finding a tree of communities, such that lower levels of the hierarchy reveal finer-grained community structures. There are two main classes of algorithms tackling this problem. Divisive (top-down) algorithms recursively partition the nodes into two communities, until a stopping rule indicates that no further split is needed. In contrast, agglomerative (bottom-up) algorithms first identify the smallest community structure and then repeatedly merge the communities using a linkage method. In this article, we establish theoretical guarantees for the recovery of the hierarchical tree and community structure of a Hierarchical Stochastic Block Model by a bottom-up algorithm. We also establish that this bottom-up algorithm attains the information-theoretic threshold for exact recovery at intermediate levels of the hierarchy. Notably, these recovery conditions are less restrictive compared to those existing for top-down algorithms. This shows that bottom-up algorithms extend the feasible region for achieving exact recovery at intermediate levels. Numerical experiments on both synthetic and real data sets confirm the superiority of bottom-up algorithms over top-down algorithms. We also observe that top-down algorithms can produce dendrograms with inversions. These findings contribute to a better understanding of hierarchical clustering techniques and their applications in network analysis.

This paper presents the observation of four-top-quark (tttt) production in proton-proton collisions at the LHC. The analysis is performed using an integrated luminosity of 140 fb^{-1} at a centre-of-mass energy of 13 TeV collected using the ATLAS detector. Events containing two leptons with the same electric charge or at least three leptons (electrons or muons) are selected. Event kinematics are used to separate signal from background through a multivariate discriminant, and dedicated control regions are used to constrain the dominant backgrounds. The observed (expected) significance of the measured tttt signal with respect to the standard model (SM) background-only hypothesis is 6.1 (4.3) standard deviations. The tttt production cross section is measured to be 22.5^{+6.6}_{-5.5} fb, consistent with the SM prediction of 12.0 pm 2.4 fb within 1.8 standard deviations. Data are also used to set limits on the three-top-quark production cross section, being an irreducible background not measured previously, and to constrain the top-Higgs Yukawa coupling and effective field theory operator coefficients that affect tttt production.

"Forward-only" algorithms, which train neural networks while avoiding a backward pass, have recently gained attention as a way of solving the biologically unrealistic aspects of backpropagation. Here, we first address compelling challenges related to the "forward-only" rules, which include reducing the performance gap with backpropagation and providing an analytical understanding of their dynamics. To this end, we show that the forward-only algorithm with top-down feedback is well-approximated by an "adaptive-feedback-alignment" algorithm, and we analytically track its performance during learning in a prototype high-dimensional setting. Then, we compare different versions of forward-only algorithms, focusing on the Forward-Forward and PEPITA frameworks, and we show that they share the same learning principles. Overall, our work unveils the connections between three key neuro-inspired learning rules, providing a link between "forward-only" algorithms, i.e., Forward-Forward and PEPITA, and an approximation of backpropagation, i.e., Feedback Alignment.

While the evaluation of explanations is an important step towards trustworthy models, it needs to be done carefully, and the employed metrics need to be well-understood. Specifically model randomization testing is often overestimated and regarded as a sole criterion for selecting or discarding certain explanation methods. To address shortcomings of this test, we start by observing an experimental gap in the ranking of explanation methods between randomization-based sanity checks [1] and model output faithfulness measures (e.g. [25]). We identify limitations of model-randomization-based sanity checks for the purpose of evaluating explanations. Firstly, we show that uninformative attribution maps created with zero pixel-wise covariance easily achieve high scores in this type of checks. Secondly, we show that top-down model randomization preserves scales of forward pass activations with high probability. That is, channels with large activations have a high probility to contribute strongly to the output, even after randomization of the network on top of them. Hence, explanations after randomization can only be expected to differ to a certain extent. This explains the observed experimental gap. In summary, these results demonstrate the inadequacy of model-randomization-based sanity checks as a criterion to rank attribution methods.

Neural HMMs are a type of neural transducer recently proposed for sequence-to-sequence modelling in text-to-speech. They combine the best features of classic statistical speech synthesis and modern neural TTS, requiring less data and fewer training updates, and are less prone to gibberish output caused by neural attention failures. In this paper, we combine neural HMM TTS with normalising flows for describing the highly non-Gaussian distribution of speech acoustics. The result is a powerful, fully probabilistic model of durations and acoustics that can be trained using exact maximum likelihood. Compared to dominant flow-based acoustic models, our approach integrates autoregression for improved modelling of long-range dependences such as utterance-level prosody. Experiments show that a system based on our proposal gives more accurate pronunciations and better subjective speech quality than comparable methods, whilst retaining the original advantages of neural HMMs. Audio examples and code are available at https://shivammehta25.github.io/OverFlow/

We introduce a simple yet effective distillation framework that is able to boost the vanilla ResNet-50 to 80%+ Top-1 accuracy on ImageNet without tricks. We construct such a framework through analyzing the problems in the existing classification system and simplify the base method ensemble knowledge distillation via discriminators by: (1) adopting the similarity loss and discriminator only on the final outputs and (2) using the average of softmax probabilities from all teacher ensembles as the stronger supervision. Intriguingly, three novel perspectives are presented for distillation: (1) weight decay can be weakened or even completely removed since the soft label also has a regularization effect; (2) using a good initialization for students is critical; and (3) one-hot/hard label is not necessary in the distillation process if the weights are well initialized. We show that such a straight-forward framework can achieve state-of-the-art results without involving any commonly-used techniques, such as architecture modification; outside training data beyond ImageNet; autoaug/randaug; cosine learning rate; mixup/cutmix training; label smoothing; etc. Our method obtains 80.67% top-1 accuracy on ImageNet using a single crop-size of 224x224 with vanilla ResNet-50, outperforming the previous state-of-the-arts by a significant margin under the same network structure. Our result can be regarded as a strong baseline using knowledge distillation, and to our best knowledge, this is also the first method that is able to boost vanilla ResNet-50 to surpass 80% on ImageNet without architecture modification or additional training data. On smaller ResNet-18, our distillation framework consistently improves from 69.76% to 73.19%, which shows tremendous practical values in real-world applications. Our code and models are available at: https://github.com/szq0214/MEAL-V2.