Part-level point cloud segmentation has recently attracted significant attention in 3D computer vision. Nevertheless, existing research is constrained by two major challenges: native 3D models lack generalization due to data scarcity, while introducing 2D pre-trained knowledge often leads to inconsistent segmentation results across different views. To address these challenges, we propose S2AM3D, which incorporates 2D segmentation priors with 3D consistent supervision. We design a point-consistent part encoder that aggregates multi-view 2D features through native 3D contrastive learning, producing globally consistent point features. A scale-aware prompt decoder is then proposed to enable real-time adjustment of segmentation granularity via continuous scale signals. Simultaneously, we introduce a large-scale, high-quality part-level point cloud dataset with more than 100k samples, providing ample supervision signals for model training. Extensive experiments demonstrate that S2AM3D achieves leading performance across multiple evaluation settings, exhibiting exceptional robustness and controllability when handling complex structures and parts with significant size variations.

Traffic signboards are vital for road safety and intelligent transportation systems, enabling navigation and autonomous driving. Yet, recognizing traffic signs at night remains underexplored due to the scarcity of realistic public datasets capturing low-light degradations and distractor classes. Existing benchmarks are predominantly daytime and do not reflect challenges such as headlight glare, motion blur, sensor noise, and vandalized or ambiguous signage. To address these gaps, we introduce INTSD, a large-scale nighttime traffic sign dataset collected across diverse regions of India. INTSD contains street-level images spanning 41 traffic signboard classes, multiple distractor categories, and varied lighting and weather conditions. The dataset is designed to support both detection and fine-grained classification under realistic nighttime scenarios. To benchmark INTSD for nighttime sign recognition, we conduct extensive evaluations using state-of-the-art detection and classification models under standardized protocols. Additionally, we present LENS-Net, a strong baseline that integrates adaptive illumination-aware detection with multimodal semantic reasoning for robust nighttime sign classification. Experiments and ablations demonstrate the challenges posed by INTSD and establish competitive baselines for future research. The dataset and code for LENS-Net is publicly available for research.

Articulated objects are ubiquitous in daily environments, and their 3D reconstruction holds great significance across various fields. However, existing articulated object reconstruction methods typically require costly inputs such as multi-stage and multi-view observations. To address the limitations, we propose a category-agnostic articulated object reconstruction framework via planar Gaussian Splatting, which only uses sparse-view RGB images from a single state. Specifically, we first introduce a Gaussian information field to perceive the optimal sparse viewpoints from candidate camera poses. To ensure precise geometric fidelity, we constrain traditional 3D Gaussians into planar primitives, facilitating accurate normal and depth estimation. The planar Gaussians are then optimized in a coarse-to-fine manner, regularized by depth smoothness and few-shot diffusion priors. Furthermore, we leverage a Vision-Language Model (VLM) via visual prompting to achieve open-vocabulary part segmentation and joint parameter estimation. Extensive experiments on both synthetic and real-world datasets demonstrate that our approach significantly outperforms existing baselines, achieving superior part-level surface reconstruction fidelity.

Acting in human environments is a crucial capability for general-purpose robots, necessitating a robust understanding of natural language and its application to physical tasks. This paper seeks to harness the capabilities of diffusion models within a visuomotor policy framework that merges visual and textual inputs to generate precise robotic trajectories. By employing reference demonstrations during training, the model learns to execute manipulation tasks specified through textual commands within the robot's immediate environment. The proposed research aims to extend an existing model by leveraging improved embeddings, and adapting techniques from diffusion models for image generation. We evaluate our methods on the CALVIN dataset, proving enhanced performance on various manipulation tasks and an increased long-horizon success rate when multiple tasks are executed in sequence. Our approach reinforces the usefulness of diffusion models and contributes towards general multitask manipulation.

Efficient machine learning deployment requires models that account for hardware constraints. Because binary logic gates are the fundamental primitives of digital hardware, models built directly from logic operations offer a promising path toward highly energy-efficient computation. Recent work has shown that networks of binary logic gates can be trained with gradient-based optimization and that their wiring can be learned. However, existing approaches remain limited in scalability and training efficiency. We address these challenges by treating the network connectome as a differentiable object and introducing a Top-K connectivity mechanism that enforces structured sparsity during training. Our resulting architecture, LILogicNet, substantially improves the efficiency of logic-gate networks. A model with only 8,000 gates trains on MNIST in under five minutes while achieving 98.45% test accuracy, matching the performance of state-of-the-art logic-gate models that require two orders of magnitude more gates. At larger scales, a 256,000-gate model achieves 60.98% test accuracy on CIFAR-10, surpassing prior approaches with comparable gate budgets. Because the final model is fully binarized and composed entirely of logic operations, inference incurs minimal compute overhead and maps naturally to a wide range of digital hardware platforms, enabling efficient deployment across diverse computing systems.

Humanoid robots have demonstrated strong capabilities for interacting with static scenes across locomotion and manipulation, yet dynamic real-world interactions remain challenging. As a step toward fast-moving object interactions, we present a reinforcement-learning training pipeline that yields a unified whole-body controller for humanoid badminton, coordinating footwork and striking without motion priors or expert demonstrations. Training follows a three-stage curriculum (footwork acquisition, precision-guided swing generation, and task-focused refinement) so legs and arms jointly serve the hitting objective. For deployment, we use an Extended Kalman Filter (EKF) to estimate and predict shuttlecock trajectories for target striking, and also develop a prediction-free variant that removes the EKF and explicit prediction. We validate the framework with five sets of experiments in simulation and on hardware. In simulation, two robots sustain a rally of 21 consecutive hits. In real-world tests with both machine-fed shuttles and human-robot rallies, the robot achieves outgoing shuttle speeds up to 19.1~m/s with a mean return landing distance of 4~m. Moreover, the prediction-free variant attains comparable performance to the EKF-based target-known policy. Overall, our approach enables dynamic yet precise goal striking in humanoid badminton and suggests a path toward more dynamics-critical whole-body interaction tasks.

Talking Face Generation (TFG) methods based on Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have recently achieved impressive progress in personalized talking head synthesis. However, existing methods typically require several minutes of reference video for meticulous preprocessing and fitting, resulting in hours of preparation time and limiting their practical applicability. In this paper, we revisit a fundamental yet underexplored question: do high-quality personalized TFG models truly require minutes-long reference videos? Our exploratory study reveals that a carefully selected reference segment of only a few seconds can often achieve performance comparable to that of using the full reference video. This finding suggests that the informativeness of reference data is more critical than its duration. Motivated by this observation, we propose ISExplore (Informative Segment Explore), a simple yet effective segment selection strategy that automatically identifies the most informative short reference segment based on three key data quality dimensions: audio feature diversity, lip movement amplitude, and viewpoint diversity. Extensive experiments demonstrate that ISExplore reduces data processing and training time by over 5x for both NeRF- and 3DGS-based methods, while preserving high-fidelity generation quality. Our method provides a practical and efficient solution for personalized TFG and offers new insights into data efficiency in 3D talking face generation.

Large Language Models (LLMs) and Vision Language Models (VLMs) have become popular tools for embodied high-level planning. However, their deployment in black-box settings often leads to unpredictable or costly errors. To harness their capabilities more reliably in robotic systems, we empirically investigate practical strategies for integrating language models as closed-loop planners. Concretely, we study how the control horizon and warm-starting impact the performance of language model-based planners. We design and conduct controlled experiments to extract actionable insights, providing recommendations that can help improve the performance and robustness of language model-based embodied planning. The full implementation and experiments are available on the project website

AI is rapidly transforming journalism, but the extent of its use in published newspaper articles remains unclear. We address this gap by auditing a large-scale dataset of 186K articles from online editions of 1.5K American newspapers published in the summer of 2025. Using Pangram, a state-of-the-art AI detector, we discover that approximately 9% of newly-published articles are either partially or fully AI-generated. This AI use is unevenly distributed, appearing more frequently in smaller, local outlets, in specific topics such as weather and technology, and within certain ownership groups. We also analyze 45K opinion pieces from Washington Post, New York Times, and Wall Street Journal, finding that they are 6.4 times more likely to contain AI-generated content than news articles from the same publications, with many AI-flagged op-eds authored by prominent public figures. Despite this prevalence, we find that AI use is rarely disclosed: a manual audit of 100 AI-flagged articles found only five disclosures of AI use. Overall, our audit highlights the immediate need for greater transparency and updated editorial standards regarding the use of AI in journalism to maintain public trust.

We present MERIT, an inference-time modular framework for multimodal misinformation detection that decomposes verification into four specialized modules: visual forensics, cross-modal alignment, retrieval-augmented claim verification, and calibrated judgment. On MMFakeBench, MERIT with GPT-4o-mini achieves 81.65% F1, outperforming all reported zero-shot baselines including GPT-4V with MMD-Agent (74.0% F1). A controlled same-model evaluation confirms gains stem from architectural design: MERIT achieves 6.14 points higher misinformation recall than MMD-Agent under identical model conditions, with per-class gains of +18.0 on visual distortion and +5.33 on textual distortion. Ablation studies reveal non-overlapping module specialization, where removing any module disproportionately degrades its target category while leaving others intact. Test set evaluation on 5,000 samples confirms generalization within 0.21 F1 points of validation results. The framework operates with any instruction-following vision-language model and produces citation-linked rationales for human review.

We study counterfactual prediction under assignment bias and propose a mathematically grounded, information-theoretic approach that removes treatment-covariate dependence without adversarial training. Starting from a bound that links the counterfactual-factual risk gap to mutual information, we learn a stochastic representation Z that is predictive of outcomes while minimizing I(Z; T). We derive a tractable variational objective that upper-bounds the information term and couples it with a supervised decoder, yielding a stable, provably motivated training criterion. The framework extends naturally to dynamic settings by applying the information penalty to sequential representations at each decision time. We evaluate the method on controlled numerical simulations and a real-world clinical dataset, comparing against recent state-of-the-art balancing, reweighting, and adversarial baselines. Across metrics of likelihood, counterfactual error, and policy evaluation, our approach performs favorably while avoiding the training instabilities and tuning burden of adversarial schemes.

Multimodal large language models (MLLMs) have made rapid progress, yet their reasoning ability often lags behind strong text-only LLMs. Bridging this gap typically requires large-scale multimodal reasoning data or reinforcement learning, incurring substantial cost. An appealing alternative is parameter-space model merging between reasoning-enhanced LLMs and MLLMs, but we show that naive merging is fragile: its effectiveness varies widely across model families and can significantly degrade performance (e.g., for Qwen-based MLLMs). We propose Directional Reasoning Injection for Fine-Tuning (DRIFT), a lightweight method that transfers reasoning knowledge in the gradient space while preserving multimodal alignment. DRIFT precomputes a reasoning prior from the parameter differences between text-only reasoning experts and multimodal models, and uses it to bias gradients during supervised fine-tuning. This design retains the simplicity of standard SFT pipelines while enabling efficient and stable reasoning transfer. Experiments on multimodal reasoning benchmarks, including MathVista and MathVerse, show that DRIFT consistently outperforms naive merging and standard SFT, and matches or surpasses training-intensive methods with substantially lower data and compute.

As retrieval-augmented generation (RAG) tackles complex tasks, increasingly expanded contexts offer richer information, but at the cost of higher latency and increased cognitive load on the model. To mitigate this bottleneck, especially for intricate multi-hop questions, we introduce BRIEF-Pro. It is a universal, lightweight compressor that distills relevant evidence for a given query from retrieved documents into a concise summary for seamless integration into in-context RAG. Using seed data consisting of relatively short contexts (fewer than 1k words), BRIEF-Pro is trained to perform abstractive compression of extended contexts exceeding 10k words across a wide range of scenarios. Furthermore, BRIEF-Pro offers flexible user control over summary length by allowing users to specify the desired number of sentences. Experiments on four open-domain multi-hop question-answering datasets show that BRIEF-Pro generates more concise and relevant summaries, enhancing performance across small, large, and proprietary language models. With the 70B reader model, 32x compression by BRIEF-Pro improves QA performance by 4.67% on average over LongLLMLingua's 9x, while requiring only 23% of its computational overhead.

Marketing Mix Modeling (MMM) estimates the impact of marketing activities on business outcomes such as sales or revenue. Traditional MMM approaches rely on linear regression or Bayesian hierarchical models that assume channel independence and struggle to capture temporal dynamics and non-linear saturation. DeepCausalMMM addresses these limitations by combining deep learning, causal inference, and marketing science. It uses Gated Recurrent Units (GRUs) to learn temporal patterns (adstock, lag) while learning statistical dependencies between channels through Directed Acyclic Graph (DAG) structure with upper triangular constraints. It implements Hill equation saturation curves for diminishing returns and budget optimization. Key features: (1) data-driven hyperparameters learned from data with defaults, (2) linear mean scaling of the dependent variable, (3) configurable attribution priors with dynamic loss scaling, (4) multi-region modeling with shared and region-specific parameters, (5) robust methods including Huber loss, (6) response curve analysis.

Aligning text-to-speech (TTS) system outputs with human feedback through preference optimization has been shown to effectively improve the robustness and naturalness of language model-based TTS models. Current approaches primarily require paired desirable and undesirable samples at the utterance level. However, such pairs are often limited in TTS output data, and utterance-level formulation prevents fine-grained token-level optimization needed for accurate pronunciation alignment. In this study, we propose TKTO that eliminates the need for paired data, enabling a more data-efficient training paradigm, and directly targets token-level units, automatically providing fine-grained alignment signals without token-level annotations. TKTO improves the challenging Japanese TTS accuracy by 39% and reduces CER by 54%, automatically assigning 12.8 times stronger reward to targeted tokens.

Context utilisation, the ability of Language Models (LMs) to incorporate relevant information from the provided context when generating responses, remains largely opaque to users, who cannot determine whether models draw from parametric memory or provided context, nor identify which specific context pieces inform the response. Highlight explanations (HEs) offer a natural solution as they can point the exact context pieces and tokens that influenced model outputs. However, no existing work evaluates their effectiveness in accurately explaining context utilisation. We address this gap by introducing the first gold standard HE evaluation framework for context attribution, using controlled test cases with known ground-truth context usage, which avoids the limitations of existing indirect proxy evaluations. To demonstrate the framework's broad applicability, we evaluate four HE methods -- three established techniques and MechLight, a mechanistic interpretability approach we adapt for this task -- across four context scenarios, four datasets, and five LMs. Overall, we find that MechLight performs best across all context scenarios. However, all methods struggle with longer contexts and exhibit positional biases, pointing to fundamental challenges in explanation accuracy that require new approaches to deliver reliable context utilisation explanations at scale.

Large audio-language models (LALMs) unify speech and text processing, but their robustness in noisy real-world settings remains underexplored. We investigate how irrelevant audio, such as silence, synthetic noise, and environmental sounds, affects text reasoning tasks where audio is unnecessary. Across three text-based benchmarks, we find that even non-informative audio reduces accuracy and increases prediction volatility; the severity of interference scales with longer durations, higher amplitudes, and elevated decoding temperatures. Silence, often assumed neutral, destabilizes outputs as strongly as synthetic noise. While larger models show greater resilience, vulnerabilities persist across all evaluated systems. We further test mitigation strategies and find that prompting shows limited effectiveness, whereas self-consistency improves stability at the cost of increased computation. Our results reveal cross-modal interference as a key robustness challenge and highlight the need for efficient fusion strategies that preserve reasoning performance in the presence of irrelevant inputs.

Reinforcement learning fine-tuning (RLFT) is a dominant paradigm for improving pretrained policies for downstream tasks. These pretrained policies, trained on large datasets, produce generations with a broad range of promising but unrefined behaviors. Often, a critical failure mode of RLFT arises when policies lose this diversity and collapse into a handful of easily exploitable outputs. This convergence hinders exploration, which is essential for expanding the capabilities of the pretrained policy and for amplifying the benefits of test-time compute scaling. To address this, we introduce an objective for policy gradient methods that explicitly enforces the exploration and refinement of diverse generations, which we call a polychromic objective. We then show how proximal policy optimization (PPO) can be adapted to optimize this objective. Our method (1) employs vine sampling to collect on-policy rollouts and (2) modifies the advantage function to reflect the advantage under our new objective. Experiments on BabyAI, Minigrid, and Algorithmic Creativity show that our method improves success rates by reliably solving a larger set of environment configurations and generalizes better under large perturbations. Moreover, when given multiple attempts in pass@$k$ experiments, the policy achieves substantially higher coverage, demonstrating its ability to maintain and exploit a diverse repertoire of strategies.

Large-scale robot learning has made progress on complex manipulation tasks, yet long horizon, contact rich problems, especially those involving deformable objects, remain challenging due to inconsistent demonstration quality. We propose a stage-aware, video-based reward modeling framework that jointly predicts task stage and fine-grained progress, using natural language subtask annotations to derive consistent labels across variable-length demonstrations. This avoids the brittleness of frame index based labeling and provides stable supervision even in tasks like T-shirt folding. Our reward model is robust to demonstration variability, generalizes to out-of-distribution scenarios, and improves downstream policy training. Building on it, we introduce Reward-Aligned Behavior Cloning (RA-BC), which filters and reweights demonstrations based on reward estimates. Experiments show that our method significantly outperforms baselines in both real-world rollouts and human validation. On T-shirt folding, we achieve 83% success from the flattened state and 67% from the crumpled state, compared to 8% and 0% with vanilla BC. Overall, our results highlight reward modeling as a scalable and annotation-efficient solution for long horizon robotic manipulation. Project website: https://qianzhong-chen.github.io/sarm.github.io/

Predicting cellular responses to genetic perturbations represents a fundamental challenge in systems biology, critical for advancing therapeutic discovery and virtual cell modeling. While large language models (LLMs) show promise for biological reasoning, their application to perturbation prediction remains underexplored due to challenges in adapting them to structured experimental data. We present SynthPert, a novel method that enhances LLM performance through supervised fine-tuning on synthetic reasoning traces generated by frontier models. Using the PerturbQA benchmark, we demonstrate that our approach not only achieves state-of-the-art performance but surpasses the capabilities of the frontier model that generated the training data. Our results reveal three key insights: (1) Synthetic reasoning traces effectively distill biological knowledge even when partially inaccurate, (2) This approach enables cross-cell-type generalization with 87% accuracy on unseen RPE1 cells, and (3) Performance gains persist despite using only 2% of quality-filtered training data. This work shows the effectiveness of synthetic reasoning distillation for enhancing domain-specific reasoning in LLMs.

Vision-Language-Action (VLA) models trained via imitation learning suffer from significant performance degradation in data-scarce scenarios due to their reliance on large-scale demonstration datasets. Although reinforcement learning (RL)-based post-training has proven effective in addressing data scarcity, its application to VLA models is hindered by the non-resettable nature of real-world environments. This limitation is particularly critical in high-risk domains such as industrial automation, where interactions often induce state changes that are costly or infeasible to revert. Furthermore, existing VLA approaches lack a reliable mechanism for detecting task completion, leading to redundant actions that reduce overall task success rates. To address these challenges, we propose World-Env, an RL-based post-training framework that replaces physical interaction with a low-cost world model-based virtual simulator. World-Env consists of two key components: (1) a physically-consistent world simulator that generates temporally consistent future visual observations, and (2) a vision-language model (VLM)-guided instant reflector that provides continuous reward signals and predicts action termination. This simulated environment enables VLA models to safely explore and generalize beyond their initial imitation learning distribution. Our method achieves notable performance gains with as few as five expert demonstrations per task. Experiments on complex robotic manipulation tasks demonstrate that World-Env effectively overcomes the data inefficiency, safety constraints, and inefficient execution of conventional VLA models that rely on real-world interaction, offering a practical and scalable solution for post-training in resource-constrained settings. Our code is available at https://github.com/amap-cvlab/world-env.

Multi-Hop Question Answering (MHQA) requires integrating dispersed, interdependent evidence through sequential reasoning under noise. This task is challenging for LLMs as they have a finite per-pass output capacity, beyond which the integration of task-relevant evidence proves unreliable. Consequently, the single-pass reasoning paradigm is inherently vulnerable to this capacity overflow. To formalize this bottleneck, our analysis establishes a Fano-style accuracy upper bound, defining a theoretical performance ceiling for single-pass LLMs. This bound reveals that accuracy inevitably collapses once task complexity exceeds model capacity, providing general principles for capacity-aware representation and structuring of MHQA in LLMs. Building on these principles, we introduce a proof-of-concept multi-call framework for MHQA, InfoQA. It ensures high per-step accuracy by combining capacity-aware task decomposition with active pruning of prior reasoning traces, keeping the information load within the single-pass limit. It further achieves robustness by a dependency-explicit workflow that enables precise control over the reasoning path. We construct a stringent and noise-rich benchmark to validate our theory and framework. Experimental results show that model behavior aligns with our predicted capacity curves while InfoQA achieves consistent performance improvements. We hope our work inspires more LLM multi-step reasoning methods: \faGithub \href{https://github.com/KaiyangWan/InfoQA}{InfoQA}.

Large Language Models (LLMs) have demonstrated strong performance across general NLP tasks, but their utility in automating numerical experiments of complex physical system -- a critical and labor-intensive component -- remains underexplored. As the major workhorse of computational science over the past decades, Computational Fluid Dynamics (CFD) offers a uniquely challenging testbed for evaluating the scientific capabilities of LLMs. We introduce CFDLLMBench, a benchmark suite comprising three complementary components -- CFDQuery, CFDCodeBench, and FoamBench -- designed to holistically evaluate LLM performance across three key competencies: graduate-level CFD knowledge, numerical and physical reasoning of CFD, and context-dependent implementation of CFD workflows. Grounded in real-world CFD practices, our benchmark combines a detailed task taxonomy with a rigorous evaluation framework to deliver reproducible results and quantify LLM performance across code executability, solution accuracy, and numerical convergence behavior. CFDLLMBench establishes a solid foundation for the development and evaluation of LLM-driven automation of numerical experiments for complex physical systems. Code and data are available at https://github.com/NREL-Theseus/cfdllmbench/.

In several applications, including in synthetic biology, one often has input/output data on a system composed of many modules, and although the modules' input/output functions and signals may be unknown, knowledge of the composition architecture can significantly reduce the amount of training data required to learn the system's input/output mapping. Learning the modules' input/output functions is also necessary for designing new systems from different composition architectures. Here, we propose a modular learning framework, which incorporates prior knowledge of the system's compositional structure to (a) identify the composing modules' input/output functions from the system's input/output data and (b) achieve this by using a reduced amount of data compared to what would be required without knowledge of the compositional structure. To achieve this, we introduce the notion of modular identifiability, which allows recovery of modules' input/output functions from a subset of the system's input/output data, and provide theoretical guarantees on a class of systems motivated by genetic circuits. We demonstrate the theory on computational studies showing that a neural network (NNET) that accounts for the compositional structure can learn the composing modules' input/output functions and predict the system's output on inputs outside of the training set distribution. By contrast, a neural network that is agnostic of the structure is unable to predict on inputs that fall outside of the training set distribution. By reducing the need for experimental data and allowing module identification, this framework offers the potential to ease the design of synthetic biological circuits and of multi-module systems more generally.

Personalized animal image generation is challenging due to rich appearance cues and large morphological variability. Existing approaches often exhibit feature misalignment across domains, which leads to identity drift. We present AnimalBooth, a framework that strengthens identity preservation with an Animal Net and an adaptive attention module, mitigating cross domain alignment errors. We further introduce a frequency controlled feature integration module that applies Discrete Cosine Transform filtering in the latent space to guide the diffusion process, enabling a coarse to fine progression from global structure to detailed texture. To advance research in this area, we curate AnimalBench, a high resolution dataset for animal personalization. Extensive experiments show that AnimalBooth consistently outperforms strong baselines on multiple benchmarks and improves both identity fidelity and perceptual quality.

Understanding and reasoning about entire software repositories is an essential capability for intelligent software engineering tools. While existing benchmarks such as CoSQA and CodeQA have advanced the field, they predominantly focus on small, self-contained code snippets. These setups fail to capture the complexity of real-world repositories, where effective understanding and reasoning often require navigating multiple files, understanding software architecture, and grounding answers in long-range code dependencies. In this paper, we present SWE-QA, a repository-level code question answering (QA) benchmark designed to facilitate research on automated QA systems in realistic code environments. SWE-QA involves 576 high-quality question-answer pairs spanning diverse categories, including intention understanding, cross-file reasoning, and multi-hop dependency analysis. To construct SWE-QA, we first crawled 77,100 GitHub issues from 11 popular repositories. Based on an analysis of naturally occurring developer questions extracted from these issues, we developed a two-level taxonomy of repository-level questions and constructed a set of seed questions for each category. For each category, we manually curated and validated questions and collected their corresponding answers. As a prototype application, we further develop SWE-QA-Agent, an agentic framework in which LLM agents reason and act to find answers automatically. We evaluate six advanced LLMs on SWE-QA under various context augmentation strategies. Experimental results highlight the promise of LLMs, particularly our SWE-QA-Agent framework, in addressing repository-level QA, while also revealing open challenges and pointing to future research directions.

Large language model based multi-agent systems (MAS) have unlocked significant advancements in tackling complex problems, but their increasing capability introduces a structural fragility that makes them difficult to debug. A key obstacle to improving their reliability is the severe scarcity of large-scale, diverse datasets for error attribution, as existing resources rely on costly and unscalable manual annotation. To address this bottleneck, we introduce Aegis, a novel framework for Automated error generation and attribution for multi-agent systems. Aegis constructs a large dataset of 9,533 trajectories with annotated faulty agents and error modes, covering diverse MAS architectures and task domains. This is achieved using a LLM-based manipulator that can adaptively inject context-aware errors into successful execution trajectories. Leveraging fine-grained labels and the structured arrangement of positive-negative sample pairs, Aegis supports three different learning paradigms: Supervised Fine-Tuning, Reinforcement Learning, and Contrastive Learning. We develop learning methods for each paradigm. Comprehensive experiments show that trained models consistently achieve substantial improvements in error attribution. Notably, several of our fine-tuned LLMs demonstrate performance competitive with or superior to proprietary models an order of magnitude larger, validating our automated data generation framework as a crucial resource for developing more robust and interpretable multi-agent systems. Our project website is available at https://kfq20.github.io/Aegis-Website/.

Text-guided sound separation enables flexible audio editing, assistive listening, and open-domain source extraction, but systems such as AudioSep remain too expensive for low-latency edge or codec-mediated deployment. Existing neural audio codec separators are efficient, yet largely restricted to fixed stems or closed taxonomies. We introduce CodecSep, a prompt-driven universal sound separation framework that extracts sources directly in neural audio codec latent space. CodecSep combines a frozen DAC backbone with a lightweight FiLM-conditioned Transformer masker driven by CLAP text embeddings, enabling open-vocabulary separation while preserving codec-native efficiency. Across dnr-v2 and five open-domain benchmarks, CodecSep consistently improves over AudioSep in SI-SDR, remains competitive in ViSQOL, and achieves clear gains in human MOS-LQS. Controlled analyses show that fine-grained prompts outperform coarse labels, and that explicit latent masking is substantially more effective than decoder-style latent generation in codec space. Qualitative diagnostics show that neural audio codec latents retain source-dependent structure, which CodecSep exploits mainly through channel-wise source-conditioned modulation. CodecSep also provides a practical code-stream deployment path. When audio is transmitted as neural audio codec codes, CodecSep maps codes to embeddings, separates directly in codec space, and outputs waveforms or re-quantized codes, avoiding the decode-separate-re-encode loop. In this regime, CodecSep requires only 1.35 GMACs end-to-end: about 54 times less compute than AudioSep in the same pipeline and 25 times lower separator-only compute, with much lower latency and memory. More broadly, CodecSep offers a blueprint for codec-native downstream audio processing.

We introduce the BIR-Adapter, a parameter-efficient diffusion adapter for blind image restoration. Diffusion-based restoration methods have demonstrated promising performance in addressing this fundamental problem in computer vision, typically relying on auxiliary feature extractors or extensive fine-tuning of pre-trained models. Building on the observation that large-scale pretrained diffusion models can retain informative representations under image degradations, BIR-Adapter introduces a parameter-efficient, plug-and-play attention mechanism that substantially reduces the number of trained parameters. To further improve reliability, we adapt a sampling guidance mechanism that mitigates hallucinations during restoration. Experiments on synthetic and real-world degradations demonstrate that BIR-Adapter achieves competitive, and in several settings superior, performance compared to state-of-the-art methods while requiring up to 36x fewer trained parameters. Moreover, the adapter-based design enables integration into existing models. We validate this generality by extending a super-resolution-only diffusion model to handle additional unknown degradations, highlighting the adaptability of our approach for broader image restoration tasks.

As large language models increasingly deployed into agentic systems, existing methods face critical gaps in observing, assessing, and mitigating deployment-specific risks. We present a comprehensive, observability-driven workflow: we introduce \textbf{AgentSeer}, observability tool which decomposes agentic executions into granular \emph{action-component} graphs; we use this decomposition to rigorously quantify the gap between model-level and agent-level jailbreaking risk via cross-model validation on GPT-OSS-20B and Gemini-2.0-flash with HarmBench under single-turn and iterative-refinement attacks; we leverage action-graph risk signals to automate iterative prompt hardening against direct and iterative jailbreak attacks. Stark differences is revealed between model-level and agentic-level vulnerability profiles. Model-level evaluation reveals baseline differences: GPT-OSS-20B (39.47\% ASR) versus Gemini-2.0-flash (50.00\% ASR), with both models showing susceptibility to social engineering. However, agentic-level assessment exposes agent-specific risks invisible to traditional evaluation. We discover "agentic-only" vulnerabilities that emerge exclusively in agentic contexts, with tool-calling showing 24-60\% higher ASR across both models. Cross-model analysis reveals universal agentic patterns, where agent transfer operations as highest-risk tools, with semantic pattern revealed rather than syntactic vulnerability mechanisms. Direct attack transfer from model-level to agentic contexts shows degraded performance of successful prompts (GPT-OSS-20B: 57\% human injection ASR; Gemini-2.0-flash: 28\%), while context-aware iterative attacks successfully compromise objectives that failed at model-level, confirming systematic vulnerabilities gaps. Action-based prompt improvement substantially reduces action-averaged agentic jailbreak success on GPT-OSS-20B (direct: 45.3\%