InfoMosaic-Bench: Evaluating Multi-Source Information Seeking in Tool-Augmented Agents

Information seeking is a fundamental requirement for humans. However, existing LLM agents rely heavily on open-web search, which exposes two fundamental weaknesses: online content is noisy and unreliable, and many real-world tasks require precise, domain-specific knowledge unavailable from the web. The emergence of the Model Context Protocol (MCP) now allows agents to interface with thousands of specialized tools, seemingly resolving this limitation. Yet it remains unclear whether agents can effectively leverage such tools -- and more importantly, whether they can integrate them with general-purpose search to solve complex tasks. Therefore, we introduce InfoMosaic-Bench, the first benchmark dedicated to multi-source information seeking in tool-augmented agents. Covering six representative domains (medicine, finance, maps, video, web, and multi-domain integration), InfoMosaic-Bench requires agents to combine general-purpose search with domain-specific tools. Tasks are synthesized with InfoMosaic-Flow, a scalable pipeline that grounds task conditions in verified tool outputs, enforces cross-source dependencies, and filters out shortcut cases solvable by trivial lookup. This design guarantees both reliability and non-triviality. Experiments with 14 state-of-the-art LLM agents reveal three findings: (i) web information alone is insufficient, with GPT-5 achieving only 38.2% accuracy and 67.5% pass rate; (ii) domain tools provide selective but inconsistent benefits, improving some domains while degrading others; and (iii) 22.4% of failures arise from incorrect tool usage or selection, highlighting that current LLMs still struggle with even basic tool handling.

Ultra-Efficient Decoding for End-to-End Neural Compression and Reconstruction

Image compression and reconstruction are crucial for various digital applications. While contemporary neural compression methods achieve impressive compression rates, the adoption of such technology has been largely hindered by the complexity and large computational costs of the convolution-based decoders during data reconstruction. To address the decoder bottleneck in neural compression, we develop a new compression-reconstruction framework based on incorporating low-rank representation in an autoencoder with vector quantization. We demonstrated that performing a series of computationally efficient low-rank operations on the learned latent representation of images can efficiently reconstruct the data with high quality. Our approach dramatically reduces the computational overhead in the decoding phase of neural compression/reconstruction, essentially eliminating the decoder compute bottleneck while maintaining high fidelity of image outputs.

Enhancing Physical Consistency in Lightweight World Models

A major challenge in deploying world models is the trade-off between size and performance. Large world models can capture rich physical dynamics but require massive computing resources, making them impractical for edge devices. Small world models are easier to deploy but often struggle to learn accurate physics, leading to poor predictions. We propose the Physics-Informed BEV World Model (PIWM), a compact model designed to efficiently capture physical interactions in bird's-eye-view (BEV) representations. PIWM uses Soft Mask during training to improve dynamic object modeling and future prediction. We also introduce a simple yet effective technique, Warm Start, for inference to enhance prediction quality with a zero-shot model. Experiments show that at the same parameter scale (400M), PIWM surpasses the baseline by 60.6% in weighted overall score. Moreover, even when compared with the largest baseline model (400M), the smallest PIWM (130M Soft Mask) achieves a 7.4% higher weighted overall score with a 28% faster inference speed.

False Sense of Security: Why Probing-based Malicious Input Detection Fails to Generalize

Large Language Models (LLMs) can comply with harmful instructions, raising serious safety concerns despite their impressive capabilities. Recent work has leveraged probing-based approaches to study the separability of malicious and benign inputs in LLMs' internal representations, and researchers have proposed using such probing methods for safety detection. We systematically re-examine this paradigm. Motivated by poor out-of-distribution performance, we hypothesize that probes learn superficial patterns rather than semantic harmfulness. Through controlled experiments, we confirm this hypothesis and identify the specific patterns learned: instructional patterns and trigger words. Our investigation follows a systematic approach, progressing from demonstrating comparable performance of simple n-gram methods, to controlled experiments with semantically cleaned datasets, to detailed analysis of pattern dependencies. These results reveal a false sense of security around current probing-based approaches and highlight the need to redesign both models and evaluation protocols, for which we provide further discussions in the hope of suggesting responsible further research in this direction. We have open-sourced the project at https://github.com/WangCheng0116/Why-Probe-Fails.

Model-Task Alignment Drives Distinct RL Outcomes

Recent advances in applying reinforcement learning (RL) to large language models (LLMs) have led to substantial progress. In particular, a series of remarkable yet often counterintuitive phenomena have been reported in LLMs, exhibiting patterns not typically observed in traditional RL settings. For example, notable claims include that a single training example can match the performance achieved with an entire dataset, that the reward signal does not need to be very accurate, and that training solely with negative samples can match or even surpass sophisticated reward-based methods. However, the precise conditions under which these observations hold - and, critically, when they fail - remain unclear. In this work, we identify a key factor that differentiates RL observations: whether the pretrained model already exhibits strong Model-Task Alignment, as measured by pass@k accuracy on the evaluated task. Through a systematic and comprehensive examination of a series of counterintuitive claims, supported by rigorous experimental validation across different model architectures and task domains, our findings show that while standard RL training remains consistently robust across settings, many of these counterintuitive results arise only when the model and task already exhibit strong model-task alignment. In contrast, these techniques fail to drive substantial learning in more challenging regimes, where standard RL methods remain effective.

EndoGen: Conditional Autoregressive Endoscopic Video Generation

Endoscopic video generation is crucial for advancing medical imaging and enhancing diagnostic capabilities. However, prior efforts in this field have either focused on static images, lacking the dynamic context required for practical applications, or have relied on unconditional generation that fails to provide meaningful references for clinicians. Therefore, in this paper, we propose the first conditional endoscopic video generation framework, namely EndoGen. Specifically, we build an autoregressive model with a tailored Spatiotemporal Grid-Frame Patterning (SGP) strategy. It reformulates the learning of generating multiple frames as a grid-based image generation pattern, which effectively capitalizes the inherent global dependency modeling capabilities of autoregressive architectures. Furthermore, we propose a Semantic-Aware Token Masking (SAT) mechanism, which enhances the model's ability to produce rich and diverse content by selectively focusing on semantically meaningful regions during the generation process. Through extensive experiments, we demonstrate the effectiveness of our framework in generating high-quality, conditionally guided endoscopic content, and improves the performance of downstream task of polyp segmentation. Code released at https://www.github.com/CUHK-AIM-Group/EndoGen.

Robust Noisy Pseudo-label Learning for Semi-supervised Medical Image Segmentation Using Diffusion Model

Obtaining pixel-level annotations in the medical domain is both expensive and time-consuming, often requiring close collaboration between clinical experts and developers. Semi-supervised medical image segmentation aims to leverage limited annotated data alongside abundant unlabeled data to achieve accurate segmentation. However, existing semi-supervised methods often struggle to structure semantic distributions in the latent space due to noise introduced by pseudo-labels. In this paper, we propose a novel diffusion-based framework for semi-supervised medical image segmentation. Our method introduces a constraint into the latent structure of semantic labels during the denoising diffusion process by enforcing prototype-based contrastive consistency. Rather than explicitly delineating semantic boundaries, the model leverages class prototypes centralized semantic representations in the latent space as anchors. This strategy improves the robustness of dense predictions, particularly in the presence of noisy pseudo-labels. We also introduce a new publicly available benchmark: Multi-Object Segmentation in X-ray Angiography Videos (MOSXAV), which provides detailed, manually annotated segmentation ground truth for multiple anatomical structures in X-ray angiography videos. Extensive experiments on the EndoScapes2023 and MOSXAV datasets demonstrate that our method outperforms state-of-the-art medical image segmentation approaches under the semi-supervised learning setting. This work presents a robust and data-efficient diffusion model that offers enhanced flexibility and strong potential for a wide range of clinical applications.

SWDL: Stratum-Wise Difference Learning with Deep Laplacian Pyramid for Semi-Supervised 3D Intracranial Hemorrhage Segmentation

Recent advances in medical imaging have established deep learning-based segmentation as the predominant approach, though it typically requires large amounts of manually annotated data. However, obtaining annotations for intracranial hemorrhage (ICH) remains particularly challenging due to the tedious and costly labeling process. Semi-supervised learning (SSL) has emerged as a promising solution to address the scarcity of labeled data, especially in volumetric medical image segmentation. Unlike conventional SSL methods that primarily focus on high-confidence pseudo-labels or consistency regularization, we propose SWDL-Net, a novel SSL framework that exploits the complementary advantages of Laplacian pyramid and deep convolutional upsampling. The Laplacian pyramid excels at edge sharpening, while deep convolutions enhance detail precision through flexible feature mapping. Our framework achieves superior segmentation of lesion details and boundaries through a difference learning mechanism that effectively integrates these complementary approaches. Extensive experiments on a 271-case ICH dataset and public benchmarks demonstrate that SWDL-Net outperforms current state-of-the-art methods in scenarios with only 2% labeled data. Additional evaluations on the publicly available Brain Hemorrhage Segmentation Dataset (BHSD) with 5% labeled data further confirm the superiority of our approach. Code and data have been released at https://github.com/SIAT-CT-LAB/SWDL.

Towards a general-purpose foundation model for fMRI analysis

Functional Magnetic Resonance Imaging (fMRI) is essential for studying brain function and diagnosing neurological disorders, but current analysis methods face reproducibility and transferability issues due to complex pre-processing and task-specific models. We introduce NeuroSTORM (Neuroimaging Foundation Model with Spatial-Temporal Optimized Representation Modeling), a generalizable framework that directly learns from 4D fMRI volumes and enables efficient knowledge transfer across diverse applications. NeuroSTORM is pre-trained on 28.65 million fMRI frames (>9,000 hours) from over 50,000 subjects across multiple centers and ages 5 to 100. Using a Mamba backbone and a shifted scanning strategy, it efficiently processes full 4D volumes. We also propose a spatial-temporal optimized pre-training approach and task-specific prompt tuning to improve transferability. NeuroSTORM outperforms existing methods across five tasks: age/gender prediction, phenotype prediction, disease diagnosis, fMRI-to-image retrieval, and task-based fMRI classification. It demonstrates strong clinical utility on datasets from hospitals in the U.S., South Korea, and Australia, achieving top performance in disease diagnosis and cognitive phenotype prediction. NeuroSTORM provides a standardized, open-source foundation model to improve reproducibility and transferability in fMRI-based clinical research.

SWE-Dev: Evaluating and Training Autonomous Feature-Driven Software Development

Large Language Models (LLMs) have shown strong capability in diverse software engineering tasks, e.g. code completion, bug fixing, and document generation. However, feature-driven development (FDD), a highly prevalent real-world task that involves developing new functionalities for large, existing codebases, remains underexplored. We therefore introduce SWE-Dev, the first large-scale dataset (with 14,000 training and 500 test samples) designed to evaluate and train autonomous coding systems on real-world feature development tasks. To ensure verifiable and diverse training, SWE-Dev uniquely provides all instances with a runnable environment and its developer-authored executable unit tests. This collection not only provides high-quality data for Supervised Fine-Tuning (SFT), but also enables Reinforcement Learning (RL) by delivering accurate reward signals from executable unit tests. Our extensive evaluations on SWE-Dev, covering 17 chatbot LLMs, 10 reasoning models, and 10 Multi-Agent Systems (MAS), reveal that FDD is a profoundly challenging frontier for current AI (e.g., Claude-3.7-Sonnet achieves only 22.45\% Pass@3 on the hard test split). Crucially, we demonstrate that SWE-Dev serves as an effective platform for model improvement: fine-tuning on training set enabled a 7B model comparable to GPT-4o on \textit{hard} split, underscoring the value of its high-quality training data. Code is available here \href{https://github.com/justLittleWhite/SWE-Dev}{https://github.com/justLittleWhite/SWE-Dev}.