Large language models (LLMs) are rapidly approaching the level of proficiency in university-level symbolic mathematics required for applications in advanced science and technology. However, existing benchmarks fall short in assessing the core skills of LLMs in symbolic mathematics-such as integration, differential equations, and algebraic simplification. To address this gap, we introduce ASyMOB, a novel assessment framework focused exclusively on symbolic manipulation, featuring 17,092 unique math challenges, organized by similarity and complexity. ASyMOB enables analysis of LLM generalization capabilities by comparing performance in problems that differ by simple numerical or symbolic `perturbations'. Evaluated LLMs exhibit substantial degradation in performance for all perturbation types (up to -70.3%), suggesting reliance on memorized patterns rather than deeper understanding of symbolic math, even among models achieving high baseline accuracy. Comparing LLM performance to computer algebra systems, we identify examples where they fail while LLMs succeed, as well as problems solved only by combining both approaches. Models capable of integrated code execution yielded higher accuracy compared to their performance without code, particularly stabilizing weaker models (up to +33.1% for certain perturbation types). Notably, the most advanced models (o4-mini, Gemini 2.5 Flash) demonstrate not only high symbolic math proficiency (scoring 96.8% and 97.6% on the unperturbed set), but also remarkable robustness against perturbations, (-21.7% and -21.2% vs. average -50.4% for the other models). This may indicate a recent "phase transition" in the generalization capabilities of frontier LLMs. It remains to be seen whether the path forward lies in deeper integration with sophisticated external tools, or in developing models so capable that symbolic math systems like CAS become unnecessary.
Despite advances in large language models (LLMs) on reasoning and instruction-following benchmarks, it remains unclear whether they can reliably produce outputs aligned with a broad variety of user goals, a concept we refer to as steerability. The abundance of methods proposed to modify LLM behavior makes it unclear whether current LLMs are already steerable, or require further intervention. In particular, LLMs may exhibit (i) poor coverage, where rare user goals are underrepresented; (ii) miscalibration, where models overshoot requests; and (iii) side effects, where changes to one dimension of text inadvertently affect others. To systematically evaluate these failures, we introduce a framework based on a multi-dimensional goal space that models user goals and LLM outputs as vectors with dimensions corresponding to text attributes (e.g., reading difficulty). Applied to a text-rewriting task, we find that current LLMs struggle with steerability, as side effects are persistent. Interventions to improve steerability, such as prompt engineering, best-of-$N$ sampling, and reinforcement learning fine-tuning, have varying effectiveness, yet side effects remain problematic. Our findings suggest that even strong LLMs struggle with steerability, and existing alignment strategies may be insufficient. We open-source our steerability evaluation framework at https://github.com/MLD3/steerability.
Unmanned aerial vehicle (UAV) usage is expected to surge in the coming decade, raising the need for heightened security measures to prevent airspace violations and security threats. This study investigates deep learning approaches to UAV classification focusing on the key issue of data scarcity. To investigate this we opted to train the models using a total of 4,500 seconds of audio samples, evenly distributed across a 9-class dataset. We leveraged parameter efficient fine-tuning (PEFT) and data augmentations to mitigate the data scarcity. This paper implements and compares the use of convolutional neural networks (CNNs) and attention-based transformers. Our results show that, CNNs outperform transformers by 1-2\% accuracy, while still being more computationally efficient. These early findings, however, point to potential in using transformers models; suggesting that with more data and further optimizations they could outperform CNNs. Future works aims to upscale the dataset to better understand the trade-offs between these approaches.
Audio-driven human animation methods, such as talking head and talking body generation, have made remarkable progress in generating synchronized facial movements and appealing visual quality videos. However, existing methods primarily focus on single human animation and struggle with multi-stream audio inputs, facing incorrect binding problems between audio and persons. Additionally, they exhibit limitations in instruction-following capabilities. To solve this problem, in this paper, we propose a novel task: Multi-Person Conversational Video Generation, and introduce a new framework, MultiTalk, to address the challenges during multi-person generation. Specifically, for audio injection, we investigate several schemes and propose the Label Rotary Position Embedding (L-RoPE) method to resolve the audio and person binding problem. Furthermore, during training, we observe that partial parameter training and multi-task training are crucial for preserving the instruction-following ability of the base model. MultiTalk achieves superior performance compared to other methods on several datasets, including talking head, talking body, and multi-person datasets, demonstrating the powerful generation capabilities of our approach.
Pathology image segmentation across multiple centers encounters significant challenges due to diverse sources of heterogeneity including imaging modalities, organs, and scanning equipment, whose variability brings representation bias and impedes the development of generalizable segmentation models. In this paper, we propose PathFL, a novel multi-alignment Federated Learning framework for pathology image segmentation that addresses these challenges through three-level alignment strategies of image, feature, and model aggregation. Firstly, at the image level, a collaborative style enhancement module aligns and diversifies local data by facilitating style information exchange across clients. Secondly, at the feature level, an adaptive feature alignment module ensures implicit alignment in the representation space by infusing local features with global insights, promoting consistency across heterogeneous client features learning. Finally, at the model aggregation level, a stratified similarity aggregation strategy hierarchically aligns and aggregates models on the server, using layer-specific similarity to account for client discrepancies and enhance global generalization. Comprehensive evaluations on four sets of heterogeneous pathology image datasets, encompassing cross-source, cross-modality, cross-organ, and cross-scanner variations, validate the effectiveness of our PathFL in achieving better performance and robustness against data heterogeneity.
Improving Multi-modal Large Language Models (MLLMs) in the post-training stage typically relies on supervised fine-tuning (SFT) or reinforcement learning (RL). However, these supervised methods require expensive and manually annotated multi-modal data--an ultimately unsustainable resource. While recent efforts have explored unsupervised post-training, their methods are complex and difficult to iterate. In this work, we are the first to investigate the use of GRPO, a stable and scalable online RL algorithm, for enabling continual self-improvement without any external supervision. We propose MM-UPT, a simple yet effective framework for unsupervised post-training of MLLMs. MM-UPT builds upon GRPO, replacing traditional reward signals with a self-rewarding mechanism based on majority voting over multiple sampled responses. Our experiments demonstrate that MM-UPT significantly improves the reasoning ability of Qwen2.5-VL-7B (e.g., 66.3 %$\rightarrow$72.9 % on MathVista, 62.9 %$\rightarrow$68.7 % on We-Math), using standard dataset without ground truth labels. MM-UPT also outperforms prior unsupervised baselines and even approaches the results of supervised GRPO. Furthermore, we show that incorporating synthetic questions, generated solely by MLLM itself, can boost performance as well, highlighting a promising approach for scalable self-improvement. Overall, MM-UPT offers a new paradigm for continual, autonomous enhancement of MLLMs in the absence of external supervision. Our code is available at https://github.com/waltonfuture/MM-UPT.
This paper presents a groundbreaking approach - the first online automatic geometric calibration method for radar and camera systems. Given the significant data sparsity and measurement uncertainty in radar height data, achieving automatic calibration during system operation has long been a challenge. To address the sparsity issue, we propose a Dual-Perspective representation that gathers features from both frontal and bird's-eye views. The frontal view contains rich but sensitive height information, whereas the bird's-eye view provides robust features against height uncertainty. We thereby propose a novel Selective Fusion Mechanism to identify and fuse reliable features from both perspectives, reducing the effect of height uncertainty. Moreover, for each view, we incorporate a Multi-Modal Cross-Attention Mechanism to explicitly find location correspondences through cross-modal matching. During the training phase, we also design a Noise-Resistant Matcher to provide better supervision and enhance the robustness of the matching mechanism against sparsity and height uncertainty. Our experimental results, tested on the nuScenes dataset, demonstrate that our method significantly outperforms previous radar-camera auto-calibration methods, as well as existing state-of-the-art LiDAR-camera calibration techniques, establishing a new benchmark for future research. The code is available at https://github.com/nycu-acm/RC-AutoCalib.
Recent advancements in large language models (LLMs) have demonstrated impressive chain-of-thought reasoning capabilities, with reinforcement learning (RL) playing a crucial role in this progress. While "aha moment" patterns--where models exhibit self-correction through reflection--are often attributed to emergent properties from RL, we first demonstrate that these patterns exist in multimodal LLMs (MLLMs) prior to RL training but may not necessarily correlate with improved reasoning performance. Building on these insights, we present a comprehensive study on enhancing multimodal reasoning through a two-stage approach: (1) supervised fine-tuning (SFT) as a cold start with structured chain-of-thought reasoning patterns, followed by (2) reinforcement learning via GRPO to further refine these capabilities. Our extensive experiments show that this combined approach consistently outperforms both SFT-only and RL-only methods across challenging multimodal reasoning benchmarks. The resulting models achieve state-of-the-art performance among open-source MLLMs at both 3B and 7B scales, with our 7B model showing substantial improvements over base models (e.g., 66.3 %$\rightarrow$73.4 % on MathVista, 62.9 %$\rightarrow$70.4 % on We-Math) and our 3B model achieving performance competitive with several 7B models. Overall, this work provides practical guidance for building advanced multimodal reasoning models. Our code is available at https://github.com/waltonfuture/RL-with-Cold-Start.
Power transmission corridor hazard segmentation (PTCHS) aims to separate transmission equipment and surrounding hazards from complex background, conveying great significance to maintaining electric power transmission safety. Recently, the Segment Anything Model (SAM) has emerged as a foundational vision model and pushed the boundaries of segmentation tasks. However, SAM struggles to deal with the target objects in complex transmission corridor scenario, especially those with fine structure. In this paper, we propose ELE-SAM, adapting SAM for the PTCHS task. Technically, we develop a Context-Aware Prompt Adapter to achieve better prompt tokens via incorporating global-local features and focusing more on key regions. Subsequently, to tackle the hazard objects with fine structure in complex background, we design a High-Fidelity Mask Decoder by leveraging multi-granularity mask features and then scaling them to a higher resolution. Moreover, to train ELE-SAM and advance this field, we construct the ELE-40K benchmark, the first large-scale and real-world dataset for PTCHS including 44,094 image-mask pairs. Experimental results for ELE-40K demonstrate the superior performance that ELE-SAM outperforms the baseline model with the average 16.8% mIoU and 20.6% mBIoU performance improvement. Moreover, compared with the state-of-the-art method on HQSeg-44K, the average 2.9% mIoU and 3.8% mBIoU absolute improvements further validate the effectiveness of our method on high-quality generic object segmentation. The source code and dataset are available at https://github.com/Hhaizee/ELE-SAM.
Computer-use agents (CUAs) promise to automate complex tasks across operating systems (OS) and the web, but remain vulnerable to indirect prompt injection. Current evaluations of this threat either lack support realistic but controlled environments or ignore hybrid web-OS attack scenarios involving both interfaces. To address this, we propose RedTeamCUA, an adversarial testing framework featuring a novel hybrid sandbox that integrates a VM-based OS environment with Docker-based web platforms. Our sandbox supports key features tailored for red teaming, such as flexible adversarial scenario configuration, and a setting that decouples adversarial evaluation from navigational limitations of CUAs by initializing tests directly at the point of an adversarial injection. Using RedTeamCUA, we develop RTC-Bench, a comprehensive benchmark with 864 examples that investigate realistic, hybrid web-OS attack scenarios and fundamental security vulnerabilities. Benchmarking current frontier CUAs identifies significant vulnerabilities: Claude 3.7 Sonnet | CUA demonstrates an ASR of 42.9%, while Operator, the most secure CUA evaluated, still exhibits an ASR of 7.6%. Notably, CUAs often attempt to execute adversarial tasks with an Attempt Rate as high as 92.5%, although failing to complete them due to capability limitations. Nevertheless, we observe concerning ASRs of up to 50% in realistic end-to-end settings, with the recently released frontier Claude 4 Opus | CUA showing an alarming ASR of 48%, demonstrating that indirect prompt injection presents tangible risks for even advanced CUAs despite their capabilities and safeguards. Overall, RedTeamCUA provides an essential framework for advancing realistic, controlled, and systematic analysis of CUA vulnerabilities, highlighting the urgent need for robust defenses to indirect prompt injection prior to real-world deployment.
Recently, test-time adaptation has attracted wide interest in the context of vision-language models for image classification. However, to the best of our knowledge, the problem is completely overlooked in dense prediction tasks such as Open-Vocabulary Semantic Segmentation (OVSS). In response, we propose a novel TTA method tailored to adapting VLMs for segmentation during test time. Unlike TTA methods for image classification, our Multi-Level and Multi-Prompt (MLMP) entropy minimization integrates features from intermediate vision-encoder layers and is performed with different text-prompt templates at both the global CLS token and local pixel-wise levels. Our approach could be used as plug-and-play for any segmentation network, does not require additional training data or labels, and remains effective even with a single test sample. Furthermore, we introduce a comprehensive OVSS TTA benchmark suite, which integrates a rigorous evaluation protocol, seven segmentation datasets, and 15 common corruptions, with a total of 82 distinct test scenarios, establishing a standardized and comprehensive testbed for future TTA research in open-vocabulary segmentation. Our experiments on this suite demonstrate that our segmentation-tailored method consistently delivers significant gains over direct adoption of TTA classification baselines.
Few-shot segmentation aims to segment unseen object categories from just a handful of annotated examples. This requires mechanisms that can both identify semantically related objects across images and accurately produce segmentation masks. We note that Segment Anything 2 (SAM2), with its prompt-and-propagate mechanism, offers both strong segmentation capabilities and a built-in feature matching process. However, we show that its representations are entangled with task-specific cues optimized for object tracking, which impairs its use for tasks requiring higher level semantic understanding. Our key insight is that, despite its class-agnostic pretraining, SAM2 already encodes rich semantic structure in its features. We propose SANSA (Semantically AligNed Segment Anything 2), a framework that makes this latent structure explicit, and repurposes SAM2 for few-shot segmentation through minimal task-specific modifications. SANSA achieves state-of-the-art performance on few-shot segmentation benchmarks specifically designed to assess generalization, outperforms generalist methods in the popular in-context setting, supports various prompts flexible interaction via points, boxes, or scribbles, and remains significantly faster and more compact than prior approaches. Code is available at https://github.com/ClaudiaCuttano/SANSA.
Despite remarkable advances made in all-in-one image restoration (AIR) for handling different types of degradations simultaneously, existing methods remain vulnerable to out-of-distribution degradations and images, limiting their real-world applicability. In this paper, we propose a multi-source representation learning framework BaryIR, which decomposes the latent space of multi-source degraded images into a continuous barycenter space for unified feature encoding and source-specific subspaces for specific semantic encoding. Specifically, we seek the multi-source unified representation by introducing a multi-source latent optimal transport barycenter problem, in which a continuous barycenter map is learned to transport the latent representations to the barycenter space. The transport cost is designed such that the representations from source-specific subspaces are contrasted with each other while maintaining orthogonality to those from the barycenter space. This enables BaryIR to learn compact representations with unified degradation-agnostic information from the barycenter space, as well as degradation-specific semantics from source-specific subspaces, capturing the inherent geometry of multi-source data manifold for generalizable AIR. Extensive experiments demonstrate that BaryIR achieves competitive performance compared to state-of-the-art all-in-one methods. Particularly, BaryIR exhibits superior generalization ability to real-world data and unseen degradations. The code will be publicly available at https://github.com/xl-tang3/BaryIR.
Self-supervised topological deep learning (TDL) represents a nascent but underexplored area with significant potential for modeling higher-order interactions in simplicial complexes and cellular complexes to derive representations of unlabeled graphs. Compared to simplicial complexes, cellular complexes exhibit greater expressive power. However, the advancement in self-supervised learning for cellular TDL is largely hindered by two core challenges: \textit{extrinsic structural constraints} inherent to cellular complexes, and intrinsic semantic redundancy in cellular representations. The first challenge highlights that traditional graph augmentation techniques may compromise the integrity of higher-order cellular interactions, while the second underscores that topological redundancy in cellular complexes potentially diminish task-relevant information. To address these issues, we introduce Cellular Complex Contrastive Learning with Adaptive Trimming (CellCLAT), a twofold framework designed to adhere to the combinatorial constraints of cellular complexes while mitigating informational redundancy. Specifically, we propose a parameter perturbation-based augmentation method that injects controlled noise into cellular interactions without altering the underlying cellular structures, thereby preserving cellular topology during contrastive learning. Additionally, a cellular trimming scheduler is employed to mask gradient contributions from task-irrelevant cells through a bi-level meta-learning approach, effectively removing redundant topological elements while maintaining critical higher-order semantics. We provide theoretical justification and empirical validation to demonstrate that CellCLAT achieves substantial improvements over existing self-supervised graph learning methods, marking a significant attempt in this domain.
Multimodal large language models (MLLMs) remain vulnerable to transferable adversarial examples. While existing methods typically achieve targeted attacks by aligning global features-such as CLIP's [CLS] token-between adversarial and target samples, they often overlook the rich local information encoded in patch tokens. This leads to suboptimal alignment and limited transferability, particularly for closed-source models. To address this limitation, we propose a targeted transferable adversarial attack method based on feature optimal alignment, called FOA-Attack, to improve adversarial transfer capability. Specifically, at the global level, we introduce a global feature loss based on cosine similarity to align the coarse-grained features of adversarial samples with those of target samples. At the local level, given the rich local representations within Transformers, we leverage clustering techniques to extract compact local patterns to alleviate redundant local features. We then formulate local feature alignment between adversarial and target samples as an optimal transport (OT) problem and propose a local clustering optimal transport loss to refine fine-grained feature alignment. Additionally, we propose a dynamic ensemble model weighting strategy to adaptively balance the influence of multiple models during adversarial example generation, thereby further improving transferability. Extensive experiments across various models demonstrate the superiority of the proposed method, outperforming state-of-the-art methods, especially in transferring to closed-source MLLMs. The code is released at https://github.com/jiaxiaojunQAQ/FOA-Attack.
Large Language Models (LLMs) have recently emerged as effective surrogate models and candidate generators within global optimization frameworks for expensive blackbox functions. Despite promising results, LLM-based methods often struggle in high-dimensional search spaces or when lacking domain-specific priors, leading to sparse or uninformative suggestions. To overcome these limitations, we propose HOLLM, a novel global optimization algorithm that enhances LLM-driven sampling by partitioning the search space into promising subregions. Each subregion acts as a ``meta-arm'' selected via a bandit-inspired scoring mechanism that effectively balances exploration and exploitation. Within each selected subregion, an LLM then proposes high-quality candidate points, without any explicit domain knowledge. Empirical evaluation on standard optimization benchmarks shows that HOLLM consistently matches or surpasses leading Bayesian optimization and trust-region methods, while substantially outperforming global LLM-based sampling strategies.
Advancing code reasoning in large language models (LLMs) is fundamentally limited by the scarcity of high-difficulty datasets, especially those with verifiable input-output test cases necessary for rigorous solution validation at scale. We introduce rStar-Coder, which significantly improves LLM code reasoning capabilities by constructing a large-scale, verified dataset of 418K competition-level code problems, 580K long-reasoning solutions along with rich test cases of varying difficulty. This is achieved through three core contributions: (1) we curate competitive programming code problems and oracle solutions to synthesize new, solvable problems; (2) we introduce a reliable input-output test case synthesis pipeline that decouples the generation into a three-step input generation method and a mutual verification mechanism for effective output labeling; (3) we augment problems with high-quality, test-case-verified long-reasoning solutions. Extensive experiments on Qwen models (1.5B-14B) across various code reasoning benchmarks demonstrate the superiority of rStar-Coder dataset, achieving leading performance comparable to frontier reasoning LLMs with much smaller model sizes. On LiveCodeBench, rStar-Coder improves Qwen2.5-7B from 17.4% to an impressive 57.3%, and Qwen2.5-14B from 23.3% to 62.5%, surpassing o3-mini (low) by3.1%. On the more challenging USA Computing Olympiad, our 7B model achieves an average pass@1 accuracy of 16.15%, outperforming the frontier-level QWQ-32B. Code and the dataset will be released at https://github.com/microsoft/rStar.
In cross-silo Federated Learning (FL), client selection is critical to ensure high model performance, yet it remains challenging due to data quality decompensation, budget constraints, and incentive compatibility. As training progresses, these factors exacerbate client heterogeneity and degrade global performance. Most existing approaches treat these challenges in isolation, making jointly optimizing multiple factors difficult. To address this, we propose Shapley-Bid Reputation Optimized Federated Learning (SBRO-FL), a unified framework integrating dynamic bidding, reputation modeling, and cost-aware selection. Clients submit bids based on their perceived data quality, and their contributions are evaluated using Shapley values to quantify their marginal impact on the global model. A reputation system, inspired by prospect theory, captures historical performance while penalizing inconsistency. The client selection problem is formulated as a 0-1 integer program that maximizes reputation-weighted utility under budget constraints. Experiments on FashionMNIST, EMNIST, CIFAR-10, and SVHN datasets show that SBRO-FL improves accuracy, convergence speed, and robustness, even in adversarial and low-bid interference scenarios. Our results highlight the importance of balancing data reliability, incentive compatibility, and cost efficiency to enable scalable and trustworthy FL deployments.
Adversarial attacks pose a significant threat to machine learning models by inducing incorrect predictions through imperceptible perturbations to input data. While these attacks have been extensively studied in unstructured data like images, their application to tabular data presents new challenges. These challenges arise from the inherent heterogeneity and complex feature interdependencies in tabular data, which differ significantly from those in image data. To address these differences, it is crucial to consider imperceptibility as a key criterion specific to tabular data. Most current research focuses primarily on achieving effective adversarial attacks, often overlooking the importance of maintaining imperceptibility. To address this gap, we propose a new benchmark for adversarial attacks on tabular data that evaluates both effectiveness and imperceptibility. In this study, we assess the effectiveness and imperceptibility of five adversarial attacks across four models using eleven tabular datasets, including both mixed and numerical-only datasets. Our analysis explores how these factors interact and influence the overall performance of the attacks. We also compare the results across different dataset types to understand the broader implications of these findings. The findings from this benchmark provide valuable insights for improving the design of adversarial attack algorithms, thereby advancing the field of adversarial machine learning on tabular data.
Multi-objective optimization (MOO) exists extensively in machine learning, and aims to find a set of Pareto-optimal solutions, called the Pareto front, e.g., it is fundamental for multiple avenues of research in federated learning (FL). Pareto-Front Learning (PFL) is a powerful method implemented using Hypernetworks (PHNs) to approximate the Pareto front. This method enables the acquisition of a mapping function from a given preference vector to the solutions on the Pareto front. However, most existing PFL approaches still face two challenges: (a) sampling rays in high-dimensional spaces; (b) failing to cover the entire Pareto Front which has a convex shape. Here, we introduce a novel PFL framework, called as PHN-HVVS, which decomposes the design space into Voronoi grids and deploys a genetic algorithm (GA) for Voronoi grid partitioning within high-dimensional space. We put forward a new loss function, which effectively contributes to more extensive coverage of the resultant Pareto front and maximizes the HV Indicator. Experimental results on multiple MOO machine learning tasks demonstrate that PHN-HVVS outperforms the baselines significantly in generating Pareto front. Also, we illustrate that PHN-HVVS advances the methodologies of several recent problems in the FL field. The code is available at https://github.com/buptcmm/phnhvvs}{https://github.com/buptcmm/phnhvvs.