Yu Pei

Shifat Sahariar Bhuiyan, Abhishek Tiwari, Yu Pei et al.

Automated program repair (APR) techniques have achieved conspicuous progress, and are now capable of producing genuinely correct fixes in scenarios that were well beyond their capabilities only a few years ago. Nevertheless, even when an APR technique can find a correct fix for a bug, it still runs the risk of ranking the fix lower than other patches that are plausible (they pass all available tests) but incorrect. This can seriously hurt the technique's practical effectiveness, as the user will have to peruse a larger number of patches before finding the correct one. This paper presents PrevaRank, a technique that ranks plausible patches produced by any APR technique according to their feature similarity with historic programmer-written fixes for similar bugs. PrevaRank implements simple heuristics, which help make it scalable and applicable to any APR tool that produces plausible patches. In our experimental evaluation, after training PrevaRank on the fix history of 81 open-source Java projects, we used it to rank patches produced by 8 Java APR tools on 168 Defects4J bugs. PrevaRank consistently improved the ranking of correct fixes: for example, it ranked a correct fix within the top-3 positions in 27% more cases than the original tools did. Other experimental results indicate that PrevaRank works robustly with a variety of APR tools and bugs, with negligible overhead.

Zhicheng Cai, Xinyuan Guo, Yu Pei et al.

Autonomous agents driven by Large Language Models (LLMs) have revolutionized reasoning and problem-solving but remain static after training, unable to grow with experience as intelligent beings do during deployment. We introduce Forward Learning with EXperience (FLEX), a gradient-free learning paradigm that enables LLM agents to continuously evolve through accumulated experience. Specifically, FLEX cultivates scalable and inheritable evolution by constructing a structured experience library through continual reflection on successes and failures during interaction with the environment. FLEX delivers substantial improvements on mathematical reasoning, chemical retrosynthesis, and protein fitness prediction (up to 23% on AIME25, 10% on USPTO50k, and 14% on ProteinGym). We further identify a clear scaling law of experiential growth and the phenomenon of experience inheritance across agents, marking a step toward scalable and inheritable continuous agent evolution. Project Page: https://flex-gensi-thuair.github.io.

Yuxuan Song, Zhe Zhang, Yu Pei et al.

Generative modeling of discrete variables is challenging yet crucial for applications in natural language processing and biological sequence design. We introduce the Shortlisting Model (SLM), a novel simplex-based diffusion model inspired by progressive candidate pruning. SLM operates on simplex centroids, reducing generation complexity and enhancing scalability. Additionally, SLM incorporates a flexible implementation of classifier-free guidance, enhancing unconditional generation performance. Extensive experiments on DNA promoter and enhancer design, protein design, character-level and large-vocabulary language modeling demonstrate the competitive performance and strong potential of SLM. Our code can be found at https://github.com/GenSI-THUAIR/SLM

Zhaodong Wu, Haochen Xue, Qi Cao et al.

Thinking with Images improves fine-grained VQA for MLLMs by emphasizing visual cues. However, tool-augmented methods depend on the capacity of grounding, which remains unreliable for MLLMs. In parallel, attention-driven methods to crop the Region of Interest (ROIs) are proposed but they are constrained by (1) fragmented attention signals scattered across layers, leading to suboptimal localization and (2) relying on question- or redundant-text-conditioned attention extraction. Our analysis reveals three patterns: MLLMs may attend to the correct region yet generate incorrect coordinates, where-to-look attention is often fragmented across layers, and attention extraction is query-sensitive. Motivated by these, We propose ConFoThinking, a Consolidated-Focused-Attention-Driven Thinking framework that learns to aggregate attention into a designated intermediate layer, from which we mine and zoom in salient regions for downstream visual understanding. Moreover, we extract attention using concise semantic cues of what to look into, which mitigates the semantic noise introduced by question- or redundant-text-based attention extraction. Experiments across five VQA benchmarks demonstrate ConFoThinking significantly improves perception performance. The code, checkpoints, and dataset will be released after being accepted.

Yu Ge, Linna Xie, Zhong Li et al.

Large Language Model Powered Multi-Agent Systems (MASs) are increasingly employed to automate complex real-world problems, such as programming and scientific discovery. Despite their promising, MASs are not without their flaws. However, failure attribution in MASs - pinpointing the specific agent actions responsible for failures - remains underexplored and labor-intensive, posing significant challenges for debugging and system improvement. To bridge this gap, we propose FAMAS, the first spectrum-based failure attribution approach for MASs, which operates through systematic trajectory replay and abstraction, followed by spectrum analysis.The core idea of FAMAS is to estimate, from variations across repeated MAS executions, the likelihood that each agent action is responsible for the failure. In particular, we propose a novel suspiciousness formula tailored to MASs, which integrates two key factor groups, namely the agent behavior group and the action behavior group, to account for the agent activation patterns and the action activation patterns within the execution trajectories of MASs. Through expensive evaluations against 12 baselines on the Who and When benchmark, FAMAS demonstrates superior performance by outperforming all the methods in comparison.

Esha Choukse, Brijesh Warrier, Scot Heath et al.

Large Artificial Intelligence (AI) training workloads spanning several tens of thousands of GPUs present unique power management challenges. These arise due to the high variability in power consumption during the training. Given the synchronous nature of these jobs, during every iteration there is a computation-heavy phase, where each GPU works on the local data, and a communication-heavy phase where all the GPUs synchronize on the data. Because compute-heavy phases require much more power than communication phases, large power swings occur. The amplitude of these power swings is ever increasing with the increase in the size of training jobs. An even bigger challenge arises from the frequency spectrum of these power swings which, if harmonized with critical frequencies of utilities, can cause physical damage to the power grid infrastructure. Therefore, to continue scaling AI training workloads safely, we need to stabilize the power of such workloads. This paper introduces the challenge with production data and explores innovative solutions across the stack: software, GPU hardware, and datacenter infrastructure. We present the pros and cons of each of these approaches and finally present a multi-pronged approach to solving the challenge. The proposed solutions are rigorously tested using a combination of real hardware and Microsoft's in-house cloud power simulator, providing critical insights into the efficacy of these interventions under real-world conditions.

Changze Lv, Jiang Zhou, Siyu Long et al.

We introduce AMix-1, a powerful protein foundation model built on Bayesian Flow Networks and empowered by a systematic training methodology, encompassing pretraining scaling laws, emergent capability analysis, in-context learning mechanism, and test-time scaling algorithm. To guarantee robust scalability, we establish a predictive scaling law and reveal the progressive emergence of structural understanding via loss perspective, culminating in a strong 1.7-billion model. Building on this foundation, we devise a multiple sequence alignment (MSA)-based in-context learning strategy to unify protein design into a general framework, where AMix-1 recognizes deep evolutionary signals among MSAs and consistently generates structurally and functionally coherent proteins. This framework enables the successful design of a dramatically improved AmeR variant with an up to $50\times$ activity increase over its wild type. Pushing the boundaries of protein engineering, we further empower AMix-1 with an evolutionary test-time scaling algorithm for in silico directed evolution that delivers substantial, scalable performance gains as verification budgets are intensified, laying the groundwork for next-generation lab-in-the-loop protein design.

Chuang Ma, Shaokai Zhao, Dongdong Zhou et al.

Micro-expression recognition (MER), a critical subfield of affective computing, presents greater challenges than macro-expression recognition due to its brief duration and low intensity. While incorporating prior knowledge has been shown to enhance MER performance, existing methods predominantly rely on simplistic, singular sources of prior knowledge, failing to fully exploit multi-source information. This paper introduces the Multi-Prior Fusion Network (MPFNet), leveraging a progressive training strategy to optimize MER tasks. We propose two complementary encoders: the Generic Feature Encoder (GFE) and the Advanced Feature Encoder (AFE), both based on Inflated 3D ConvNets (I3D) with Coordinate Attention (CA) mechanisms, to improve the model's ability to capture spatiotemporal and channel-specific features. Inspired by developmental psychology, we present two variants of MPFNet--MPFNet-P and MPFNet-C--corresponding to two fundamental modes of infant cognitive development: parallel and hierarchical processing. These variants enable the evaluation of different strategies for integrating prior knowledge. Extensive experiments demonstrate that MPFNet significantly improves MER accuracy while maintaining balanced performance across categories, achieving accuracies of 0.811, 0.924, and 0.857 on the SMIC, CASME II, and SAMM datasets, respectively. To the best of our knowledge, our approach achieves state-of-the-art performance on the SMIC and SAMM datasets.

Chuang Ma, Yu Pei, Jianhang Zhang et al.

Micro-expressions (MEs) are subtle, fleeting nonverbal cues that reveal an individual's genuine emotional state. Their analysis has attracted considerable interest due to its promising applications in fields such as healthcare, criminal investigation, and human-computer interaction. However, existing ME research is limited to single visual modality, overlooking the rich emotional information conveyed by other physiological modalities, resulting in ME recognition and spotting performance far below practical application needs. Therefore, exploring the cross-modal association mechanism between ME visual features and physiological signals (PS), and developing a multimodal fusion framework, represents a pivotal step toward advancing ME analysis. This study introduces a novel ME dataset, MMME, which, for the first time, enables synchronized collection of facial action signals (MEs), central nervous system signals (EEG), and peripheral PS (PPG, RSP, SKT, EDA, and ECG). By overcoming the constraints of existing ME corpora, MMME comprises 634 MEs, 2,841 macro-expressions (MaEs), and 2,890 trials of synchronized multimodal PS, establishing a robust foundation for investigating ME neural mechanisms and conducting multimodal fusion-based analyses. Extensive experiments validate the dataset's reliability and provide benchmarks for ME analysis, demonstrating that integrating MEs with PS significantly enhances recognition and spotting performance. To the best of our knowledge, MMME is the most comprehensive ME dataset to date in terms of modality diversity. It provides critical data support for exploring the neural mechanisms of MEs and uncovering the visual-physiological synergistic effects, driving a paradigm shift in ME research from single-modality visual analysis to multimodal fusion. The dataset will be publicly available upon acceptance of this paper.

Shiyu Zhang, Haoyang Song, Qixin Wang et al.

Reinforcement Learning (RL) has gained significant attention across various domains. However, the increasing complexity of RL programs presents testing challenges, particularly the oracle problem: defining the correctness of the RL program. Conventional human oracles struggle to cope with the complexity, leading to inefficiencies and potential unreliability in RL testing. To alleviate this problem, we propose an automated oracle approach that leverages RL properties using fuzzy logic. Our oracle quantifies an agent's behavioral compliance with reward policies and analyzes its trend over training episodes. It labels an RL program as "Buggy" if the compliance trend violates expectations derived from RL characteristics. We evaluate our oracle on RL programs with varying complexities and compare it with human oracles. Results show that while human oracles perform well in simpler testing scenarios, our fuzzy oracle demonstrates superior performance in complex environments. The proposed approach shows promise in addressing the oracle problem for RL testing, particularly in complex cases where manual testing falls short. It offers a potential solution to improve the efficiency, reliability, and scalability of RL program testing. This research takes a step towards automated testing of RL programs and highlights the potential of fuzzy logic-based oracles in tackling the oracle problem.

Yuxiang Zhu, Minxue Pan, Yu Pei et al.

More and more users and developers are using Issue Tracking Systems (ITSs) to report issues, including bugs, feature requests, enhancement suggestions, etc. Different information, however, is gathered from users when issues are reported on different ITSs, which presents considerable challenges for issue classification tools to work effectively across the ITSs. Besides, bugs often take higher priority when it comes to classifying the issues, while existing approaches to issue classification seldom focus on distinguishing bugs and the other non-bug issues, leading to suboptimal accuracy in bug identification. In this paper, we propose a deep learning-based approach to automatically identify bug-reporting issues across various ITSs. The approach implements the k-NN algorithm to detect and correct misclassifications in data extracted from the ITSs, and trains an attention-based bi-directional long short-term memory (ABLSTM) network using a dataset of over 1.2 million labelled issues to identify bug reports. Experimental evaluation shows that our approach achieved an F-measure of 85.6\% in distinguishing bugs and other issues, significantly outperforming the other benchmark and state-of-the-art approaches examined in the experiment.

Tongtong Xu, Liushan Chen, Yu Pei et al.

Fault localization is a crucial step of automated program repair, because accurately identifying program locations that are most closely implicated with a fault greatly affects the effectiveness of the patching process. An ideal fault localization technique would provide precise information while requiring moderate computational resources---to best support an efficient search for correct fixes. In contrast, most automated program repair tools use standard fault localization techniques---which are not tightly integrated with the overall program repair process, and hence deliver only subpar efficiency. In this paper, we present retrospective fault localization: a novel fault localization technique geared to the requirements of automated program repair. A key idea of retrospective fault localization is to reuse the outcome of failed patch validation to support mutation-based dynamic analysis---providing accurate fault localization information without incurring onerous computational costs. We implemented retrospective fault localization in a tool called RESTORE---based on the JAID Java program repair system. Experiments involving faults from the Defects4J standard benchmark indicate that retrospective fault localization can boost automated program repair: RESTORE efficiently explores a large fix space, delivering state-of-the-art effectiveness (41 Defects4J bugs correctly fixed, 8 more than any other automated repair tools for Java) while simultaneously boosting performance (speedup over 3 compared to JAID). Retrospective fault localization is applicable to any automated program repair techniques that rely on fault localization and dynamic validation of patches.

Yu Pei, Carlo A. Furia, Martin Nordio et al.

This paper describes AutoFix, an automatic debugging technique that can fix faults in general-purpose software. To provide high-quality fix suggestions and to enable automation of the whole debugging process, AutoFix relies on the presence of simple specification elements in the form of contracts (such as pre- and postconditions). Using contracts enhances the precision of dynamic analysis techniques for fault detection and localization, and for validating fixes. The only required user input to the AutoFix supporting tool is then a faulty program annotated with contracts; the tool produces a collection of validated fixes for the fault ranked according to an estimate of their suitability. In an extensive experimental evaluation, we applied AutoFix to over 200 faults in four code bases of different maturity and quality (of implementation and of contracts). AutoFix successfully fixed 42% of the faults, producing, in the majority of cases, corrections of quality comparable to those competent programmers would write; the used computational resources were modest, with an average time per fix below 20 minutes on commodity hardware. These figures compare favorably to the state of the art in automated program fixing, and demonstrate that the AutoFix approach is successfully applicable to reduce the debugging burden in real-world scenarios.

Nadia Polikarpova, Carlo A. Furia, Yu Pei et al.

Experience with lightweight formal methods suggests that programmers are willing to write specification if it brings tangible benefits to their usual development activities. This paper considers stronger specifications and studies whether they can be deployed as an incremental practice that brings additional benefits without being unacceptably expensive. We introduce a methodology that extends Design by Contract to write strong specifications of functional properties in the form of preconditions, postconditions, and invariants. The methodology aims at being palatable to developers who are not fluent in formal techniques but are comfortable with writing simple specifications. We evaluate the cost and the benefits of using strong specifications by applying the methodology to testing data structure implementations written in Eiffel and C#. In our extensive experiments, testing against strong specifications detects twice as many bugs as standard contracts, with a reasonable overhead in terms of annotation burden and run-time performance while testing. In the wide spectrum of formal techniques for software quality, testing against strong specifications lies in a "sweet spot" with a favorable benefit to effort ratio.