Vision-language models (VLMs) exhibit strong zero-shot generalization on natural images and show early promise in interpretable medical image analysis. However, existing benchmarks do not systematically evaluate whether these models truly reason like human clinicians or merely imitate superficial patterns. To address this gap, we propose DrVD-Bench, the first multimodal benchmark for clinical visual reasoning. DrVD-Bench consists of three modules: Visual Evidence Comprehension, Reasoning Trajectory Assessment, and Report Generation Evaluation, comprising a total of 7,789 image-question pairs. Our benchmark covers 20 task types, 17 diagnostic categories, and five imaging modalities-CT, MRI, ultrasound, radiography, and pathology. DrVD-Bench is explicitly structured to reflect the clinical reasoning workflow from modality recognition to lesion identification and diagnosis. We benchmark 19 VLMs, including general-purpose and medical-specific, open-source and proprietary models, and observe that performance drops sharply as reasoning complexity increases. While some models begin to exhibit traces of human-like reasoning, they often still rely on shortcut correlations rather than grounded visual understanding. DrVD-Bench offers a rigorous and structured evaluation framework to guide the development of clinically trustworthy VLMs.
Recent advances in deep learning-based medical image registration have shown that training deep neural networks~(DNNs) does not necessarily require medical images. Previous work showed that DNNs trained on randomly generated images with carefully designed noise and contrast properties can still generalize well to unseen medical data. Building on this insight, we propose using registration between random images as a proxy task for pretraining a foundation model for image registration. Empirical results show that our pretraining strategy improves registration accuracy, reduces the amount of domain-specific data needed to achieve competitive performance, and accelerates convergence during downstream training, thereby enhancing computational efficiency.
In this work, we focus on the task of weakly supervised affordance grounding, where a model is trained to identify affordance regions on objects using human-object interaction images and egocentric object images without dense labels. Previous works are mostly built upon class activation maps, which are effective for semantic segmentation but may not be suitable for locating actions and functions. Leveraging recent advanced foundation models, we develop a supervised training pipeline based on pseudo labels. The pseudo labels are generated from an off-the-shelf part segmentation model, guided by a mapping from affordance to part names. Furthermore, we introduce three key enhancements to the baseline model: a label refining stage, a fine-grained feature alignment process, and a lightweight reasoning module. These techniques harness the semantic knowledge of static objects embedded in off-the-shelf foundation models to improve affordance learning, effectively bridging the gap between objects and actions. Extensive experiments demonstrate that the performance of the proposed model has achieved a breakthrough improvement over existing methods. Our codes are available at https://github.com/woyut/WSAG-PLSP .
A long-held challenge in no-reference image quality assessment (NR-IQA) learning from human subjective perception is the lack of objective generalization to unseen natural distortions. To address this, we integrate a novel Depth-Guided cross-attention and refinement (Depth-CAR) mechanism, which distills scene depth and spatial features into a structure-aware representation for improved NR-IQA. This brings in the knowledge of object saliency and relative contrast of the scene for more discriminative feature learning. Additionally, we introduce the idea of TCB (Transformer-CNN Bridge) to fuse high-level global contextual dependencies from a transformer backbone with local spatial features captured by a set of hierarchical CNN (convolutional neural network) layers. We implement TCB and Depth-CAR as multimodal attention-based projection functions to select the most informative features, which also improve training time and inference efficiency. Experimental results demonstrate that our proposed DGIQA model achieves state-of-the-art (SOTA) performance on both synthetic and authentic benchmark datasets. More importantly, DGIQA outperforms SOTA models on cross-dataset evaluations as well as in assessing natural image distortions such as low-light effects, hazy conditions, and lens flares.
Although Alzheimer's disease detection via MRIs has advanced significantly thanks to contemporary deep learning models, challenges such as class imbalance, protocol variations, and limited dataset diversity often hinder their generalization capacity. To address this issue, this article focuses on the single domain generalization setting, where given the data of one domain, a model is designed and developed with maximal performance w.r.t. an unseen domain of distinct distribution. Since brain morphology is known to play a crucial role in Alzheimer's diagnosis, we propose the use of learnable pseudo-morphological modules aimed at producing shape-aware, anatomically meaningful class-specific augmentations in combination with a supervised contrastive learning module to extract robust class-specific representations. Experiments conducted across three datasets show improved performance and generalization capacity, especially under class imbalance and imaging protocol variations. The source code will be made available upon acceptance at https://github.com/zobia111/SDG-Alzheimer.
Existing long-term video prediction methods often rely on an autoregressive video prediction mechanism. However, this approach suffers from error propagation, particularly in distant future frames. To address this limitation, this paper proposes the first AutoRegression-Free (ARFree) video prediction framework using diffusion models. Different from an autoregressive video prediction mechanism, ARFree directly predicts any future frame tuples from the context frame tuple. The proposed ARFree consists of two key components: 1) a motion prediction module that predicts a future motion using motion feature extracted from the context frame tuple; 2) a training method that improves motion continuity and contextual consistency between adjacent future frame tuples. Our experiments with two benchmark datasets show that the proposed ARFree video prediction framework outperforms several state-of-the-art video prediction methods.
Recent advances in text-to-video (T2V) diffusion models have significantly enhanced the quality of generated videos. However, their ability to produce explicit or harmful content raises concerns about misuse and potential rights violations. Inspired by the success of unlearning techniques in erasing undesirable concepts from text-to-image (T2I) models, we extend unlearning to T2V models and propose a robust and precise unlearning method. Specifically, we adopt negatively-guided velocity prediction fine-tuning and enhance it with prompt augmentation to ensure robustness against LLM-refined prompts. To achieve precise unlearning, we incorporate a localization and a preservation regularization to preserve the model's ability to generate non-target concepts. Extensive experiments demonstrate that our method effectively erases a specific concept while preserving the model's generation capability for all other concepts, outperforming existing methods. We provide the unlearned models in \href{https://github.com/VDIGPKU/T2VUnlearning.git}{https://github.com/VDIGPKU/T2VUnlearning.git}.
Recent advances in automated theorem proving (ATP) through LLMs have highlighted the potential of formal reasoning with Lean 4 codes. However, ATP has not yet be revolutionized by the recent posttraining scaling as demonstrated by Open AI O1/O3 and Deepseek R1. In this work, we investigate the entire posttraining of ATP, aiming to align it with breakthroughs in reasoning models in natural languages. To begin, we continual train current ATP models with a hybrid dataset, which consists of numerous statement-proof pairs, and additional data aimed at incorporating cognitive behaviors that emulate human reasoning and hypothesis refinement. Next, we explore reinforcement learning with the use of outcome reward returned by Lean 4 compiler. Through our designed continual training and reinforcement learning processes, we have successfully improved existing formal provers, including both DeepSeek-Prover-v1.5 and Goedel-Prover, achieving state-of-the-art performance in the field of whole-proof generation. For example, we achieve a 59.8% pass rate (pass@32) on MiniF2F. This is an on-going project and we will progressively update our findings, release our data and training details.
Deep reasoning is fundamental for solving complex tasks, especially in vision-centric scenarios that demand sequential, multimodal understanding. However, existing benchmarks typically evaluate agents with fully synthetic, single-turn queries, limited visual modalities, and lack a framework to assess reasoning quality over multiple steps as required in real-world settings. To address this, we introduce Agent-X, a large-scale benchmark for evaluating vision-centric agents multi-step and deep reasoning capabilities in real-world, multimodal settings. Agent- X features 828 agentic tasks with authentic visual contexts, including images, multi-image comparisons, videos, and instructional text. These tasks span six major agentic environments: general visual reasoning, web browsing, security and surveillance, autonomous driving, sports, and math reasoning. Our benchmark requires agents to integrate tool use with explicit, stepwise decision-making in these diverse settings. In addition, we propose a fine-grained, step-level evaluation framework that assesses the correctness and logical coherence of each reasoning step and the effectiveness of tool usage throughout the task. Our results reveal that even the best-performing models, including GPT, Gemini, and Qwen families, struggle to solve multi-step vision tasks, achieving less than 50% full-chain success. These findings highlight key bottlenecks in current LMM reasoning and tool-use capabilities and identify future research directions in vision-centric agentic reasoning models. Our data and code are publicly available at https://github.com/mbzuai-oryx/Agent-X
As large language models (LLMs) are increasingly used in legal applications, current evaluation benchmarks tend to focus mainly on factual accuracy while largely neglecting important linguistic quality aspects such as clarity, coherence, and terminology. To address this gap, we propose three steps: First, we develop a regression model to evaluate the quality of legal texts based on clarity, coherence, and terminology. Second, we create a specialized set of legal questions. Third, we analyze 49 LLMs using this evaluation framework. Our analysis identifies three key findings: First, model quality levels off at 14 billion parameters, with only a marginal improvement of $2.7\%$ noted at 72 billion parameters. Second, engineering choices such as quantization and context length have a negligible impact, as indicated by statistical significance thresholds above 0.016. Third, reasoning models consistently outperform base architectures. A significant outcome of our research is the release of a ranking list and Pareto analysis, which highlight the Qwen3 series as the optimal choice for cost-performance tradeoffs. This work not only establishes standardized evaluation protocols for legal LLMs but also uncovers fundamental limitations in current training data refinement approaches. Code and models are available at: https://github.com/lyxx3rd/LegalEval-Q.
Recent advancements in text-to-image (T2I) generation have enabled models to produce high-quality images from textual descriptions. However, these models often struggle with complex instructions involving multiple objects, attributes, and spatial relationships. Existing benchmarks for evaluating T2I models primarily focus on general text-image alignment and fail to capture the nuanced requirements of complex, multi-faceted prompts. Given this gap, we introduce LongBench-T2I, a comprehensive benchmark specifically designed to evaluate T2I models under complex instructions. LongBench-T2I consists of 500 intricately designed prompts spanning nine diverse visual evaluation dimensions, enabling a thorough assessment of a model's ability to follow complex instructions. Beyond benchmarking, we propose an agent framework (Plan2Gen) that facilitates complex instruction-driven image generation without requiring additional model training. This framework integrates seamlessly with existing T2I models, using large language models to interpret and decompose complex prompts, thereby guiding the generation process more effectively. As existing evaluation metrics, such as CLIPScore, fail to adequately capture the nuances of complex instructions, we introduce an evaluation toolkit that automates the quality assessment of generated images using a set of multi-dimensional metrics. The data and code are released at https://github.com/yczhou001/LongBench-T2I.
A limitation of modern document retrieval embedding methods is that they typically encode passages (chunks) from the same documents independently, often overlooking crucial contextual information from the rest of the document that could greatly improve individual chunk representations. In this work, we introduce ConTEB (Context-aware Text Embedding Benchmark), a benchmark designed to evaluate retrieval models on their ability to leverage document-wide context. Our results show that state-of-the-art embedding models struggle in retrieval scenarios where context is required. To address this limitation, we propose InSeNT (In-sequence Negative Training), a novel contrastive post-training approach which combined with late chunking pooling enhances contextual representation learning while preserving computational efficiency. Our method significantly improves retrieval quality on ConTEB without sacrificing base model performance. We further find chunks embedded with our method are more robust to suboptimal chunking strategies and larger retrieval corpus sizes. We open-source all artifacts at https://github.com/illuin-tech/contextual-embeddings.
As large language models (LLMs) are increasingly used in high-stakes domains, accurately assessing their confidence is crucial. Humans typically express confidence through epistemic markers (e.g., "fairly confident") instead of numerical values. However, it remains unclear whether LLMs consistently use these markers to reflect their intrinsic confidence due to the difficulty of quantifying uncertainty associated with various markers. To address this gap, we first define marker confidence as the observed accuracy when a model employs an epistemic marker. We evaluate its stability across multiple question-answering datasets in both in-distribution and out-of-distribution settings for open-source and proprietary LLMs. Our results show that while markers generalize well within the same distribution, their confidence is inconsistent in out-of-distribution scenarios. These findings raise significant concerns about the reliability of epistemic markers for confidence estimation, underscoring the need for improved alignment between marker based confidence and actual model uncertainty. Our code is available at https://github.com/HKUST-KnowComp/MarCon.
The scientific literature is growing rapidly, making it hard to keep track of the state-of-the-art. Systematic literature reviews (SLRs) aim to identify and evaluate all relevant papers on a topic. After retrieving a set of candidate papers, the abstract screening phase determines initial relevance. To date, abstract screening methods using large language models (LLMs) focus on binary classification settings; existing question answering (QA) based ranking approaches suffer from error propagation. LLMs offer a unique opportunity to evaluate the SLR's inclusion and exclusion criteria, yet, existing benchmarks do not provide them exhaustively. We manually extract these criteria as well as research questions for 57 SLRs, mostly in the medical domain, enabling principled comparisons between approaches. Moreover, we propose LGAR, a zero-shot LLM Guided Abstract Ranker composed of an LLM based graded relevance scorer and a dense re-ranker. Our extensive experiments show that LGAR outperforms existing QA-based methods by 5-10 pp. in mean average precision. Our code and data is publicly available.
Extensively evaluating the capabilities of (large) language models is difficult. Rapid development of state-of-the-art models induce benchmark saturation, while creating more challenging datasets is labor-intensive. Inspired by the recent developments in mechanistic interpretability, we introduce circuit stability as a new way to assess model performance. Circuit stability refers to a model's ability to apply a consistent reasoning process-its circuit-across various inputs. We mathematically formalize circuit stability and circuit equivalence. Then, through three case studies, we empirically show that circuit stability and the lack thereof can characterize and predict different aspects of generalization. Our proposed methods offer a step towards rigorously relating the generality of models to their interpretability.
Multimodal Large Language Models (MLLMs) have experienced rapid development in recent years. However, in the financial domain, there is a notable lack of effective and specialized multimodal evaluation datasets. To advance the development of MLLMs in the finance domain, we introduce FinMME, encompassing more than 11,000 high-quality financial research samples across 18 financial domains and 6 asset classes, featuring 10 major chart types and 21 subtypes. We ensure data quality through 20 annotators and carefully designed validation mechanisms. Additionally, we develop FinScore, an evaluation system incorporating hallucination penalties and multi-dimensional capability assessment to provide an unbiased evaluation. Extensive experimental results demonstrate that even state-of-the-art models like GPT-4o exhibit unsatisfactory performance on FinMME, highlighting its challenging nature. The benchmark exhibits high robustness with prediction variations under different prompts remaining below 1%, demonstrating superior reliability compared to existing datasets. Our dataset and evaluation protocol are available at https://huggingface.co/datasets/luojunyu/FinMME and https://github.com/luo-junyu/FinMME.
Byte Pair Encoding (BPE) tokenizers, widely used in Large Language Models, face challenges in multilingual settings, including penalization of non-Western scripts and the creation of tokens with partial UTF-8 sequences. Pretokenization, often reliant on complex regular expressions, can also introduce fragility and unexpected edge cases. We propose SCRIPT (Script Category Representation in PreTokenization), a novel encoding scheme that bypasses UTF-8 byte conversion by using initial tokens based on Unicode script and category properties. This approach enables a simple, rule-based pretokenization strategy that respects script boundaries, offering a robust alternative to pretokenization strategies based on regular expressions. We also introduce and validate a constrained BPE merging strategy that enforces character integrity, applicable to both SCRIPT-BPE and byte-based BPE. Our experiments demonstrate that SCRIPT-BPE achieves competitive compression while eliminating encoding-based penalties for non-Latin-script languages.
Large Language Models (LLMs) excel in various natural language processing tasks but remain vulnerable to generating harmful content or being exploited for malicious purposes. Although safety alignment datasets have been introduced to mitigate such risks through supervised fine-tuning (SFT), these datasets often lack comprehensive risk coverage. Most existing datasets focus primarily on lexical diversity while neglecting other critical dimensions. To address this limitation, we propose a novel analysis framework to systematically measure the risk coverage of alignment datasets across three essential dimensions: Lexical Diversity, Malicious Intent, and Jailbreak Tactics. We further introduce TRIDENT, an automated pipeline that leverages persona-based, zero-shot LLM generation to produce diverse and comprehensive instructions spanning these dimensions. Each harmful instruction is paired with an ethically aligned response, resulting in two datasets: TRIDENT-Core, comprising 26,311 examples, and TRIDENT-Edge, with 18,773 examples. Fine-tuning Llama 3.1-8B on TRIDENT-Edge demonstrates substantial improvements, achieving an average 14.29% reduction in Harm Score, and a 20% decrease in Attack Success Rate compared to the best-performing baseline model fine-tuned on the WildBreak dataset.
Large Reasoning Models (LRMs) achieve superior performance by extending the thought length. However, a lengthy thinking trajectory leads to reduced efficiency. Most of the existing methods are stuck in the assumption of overthinking and attempt to reason efficiently by compressing the Chain-of-Thought, but this often leads to performance degradation. To address this problem, we introduce A*-Thought, an efficient tree search-based unified framework designed to identify and isolate the most essential thoughts from the extensive reasoning chains produced by these models. It formulates the reasoning process of LRMs as a search tree, where each node represents a reasoning span in the giant reasoning space. By combining the A* search algorithm with a cost function specific to the reasoning path, it can efficiently compress the chain of thought and determine a reasoning path with high information density and low cost. In addition, we also propose a bidirectional importance estimation mechanism, which further refines this search process and enhances its efficiency beyond uniform sampling. Extensive experiments on several advanced math tasks show that A*-Thought effectively balances performance and efficiency over a huge search space. Specifically, A*-Thought can improve the performance of QwQ-32B by 2.39$\times$ with low-budget and reduce the length of the output token by nearly 50% with high-budget. The proposed method is also compatible with several other LRMs, demonstrating its generalization capability. The code can be accessed at: https://github.com/AI9Stars/AStar-Thought.
Cultural Heritage (CH) data hold invaluable knowledge, reflecting the history, traditions, and identities of societies, and shaping our understanding of the past and present. However, many CH collections contain outdated or offensive descriptions that reflect historical biases. CH Institutions (CHIs) face significant challenges in curating these data due to the vast scale and complexity of the task. To address this, we develop an AI-powered tool that detects offensive terms in CH metadata and provides contextual insights into their historical background and contemporary perception. We leverage a multilingual vocabulary co-created with marginalized communities, researchers, and CH professionals, along with traditional NLP techniques and Large Language Models (LLMs). Available as a standalone web app and integrated with major CH platforms, the tool has processed over 7.9 million records, contextualizing the contentious terms detected in their metadata. Rather than erasing these terms, our approach seeks to inform, making biases visible and providing actionable insights for creating more inclusive and accessible CH collections.