Pinjia He
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Pinjia He has a total of 92 co-authors. Here are the top co-authors by number of shared publications:
- Michael R. Lyu
The Chinese University of Hong Kong
5 shared publications - Youliang Yuan
School of Data Science The Chinese University of Hong Kong Shenzhen China
5 shared publications - Wenxuan Wang
The Chinese University of Hong Kong Hong Kong China
4 shared publications - Xiaoying Tang
The Chinese University of Hong Kong
4 shared publications - Zhaopeng Tu
Shanghai Jiao Tong University
4 shared publications - Boxi Yu
The Chinese University of Hong Kong Shenzhen
3 shared publications
Latest Publications
- Open Access
SWE-Manager: Selecting and Synthesizing Golden Proposals Before Coding
Large language model (LLM) research in software engineering has largely focused on tasks such as code generation and bug repair. In practice, teams often draft multiple candidate proposals for fixing an issue and then deliberate on one golden proposal for implementation. This selection requires not only assessing the issue's scope, impact, and urgency, but also a clear understanding of each proposal's strengths and weaknesses. A good selection could make issue resolution more reliable while reducing regression and operational risk, whereas a poor choice can increase risk and even cause unpredictable failures. We first conduct a manual study of real-world issues to characterize the rationales maintainers use when selecting among competing proposals. Motivated by these findings, we introduce SWE-Manager, a joint selection and synthesis approach that selects the best proposal and synthesizes a golden proposal. SWE-Manager is an 8B model trained via reinforcement learning (RL) to compare proposals, justify its choice, and synthesize a golden proposal for implementation. We view proposal selection as a reasoning task, mirroring how technical managers review competing proposals by weighing issue context and each proposal's solution without executing code or running tests. On the SWE-Lancer Manager benchmark, SWE-Manager achieves 53.21 selection accuracy and 57.75 earn rate, earning 152,750 dollars and outperforming strong baselines including GPT-5. To further evaluate the effectiveness of SWE-Manager in real-world issue resolution, we design the P2A framework, which simulates a real-world workflow where multiple proposals are drafted, reviewed, and a golden proposal is selected for implementation ...
Open AccessSWE-Manager: Selecting and Synthesizing Golden Proposals Before Coding
Large language model (LLM) research in software engineering has largely focused on tasks such as code generation and bug repair. In practice, teams often draft multiple candidate proposals for fixing an issue and then deliberate on one golden proposal for implementation. This selection requires not only assessing the issue's scope, impact, and urgency, but also a clear understanding of each proposal's strengths and weaknesses. A good selection could make issue resolution more reliable while reducing regression and operational risk, whereas a poor choice can increase risk and even cause unpredictable failures. We first conduct a manual study of real-world issues to characterize the rationales maintainers use when selecting among competing proposals. Motivated by these findings, we introduce SWE-Manager, a joint selection and synthesis approach that selects the best proposal and synthesizes a golden proposal. SWE-Manager is an 8B model trained via reinforcement learning (RL) to compare proposals, justify its choice, and synthesize a golden proposal for implementation. We view proposal selection as a reasoning task, mirroring how technical managers review competing proposals by weighing issue context and each proposal's solution without executing code or running tests. On the SWE-Lancer Manager benchmark, SWE-Manager achieves 53.21 selection accuracy and 57.75 earn rate, earning 152,750 dollars and outperforming strong baselines including GPT-5. To further evaluate the effectiveness of SWE-Manager in real-world issue resolution, we design the P2A framework, which simulates a real-world workflow where multiple proposals are drafted, reviewed, and a golden proposal is selected for implementation ...
- Open Access
MicLog: Towards Accurate and Efficient LLM-based Log Parsing via Progressive Meta In-Context Learning
Log parsing converts semi-structured logs into structured templates, forming a critical foundation for downstream analysis. Traditional syntax and semantic-based parsers often struggle with semantic variations in evolving logs and data scarcity stemming from their limited domain coverage. Recent large language model (LLM)-based parsers leverage in-context learning (ICL) to extract semantics from examples, demonstrating superior accuracy. However, LLM-based parsers face two main challenges: 1) underutilization of ICL capabilities, particularly in dynamic example selection and cross-domain generalization, leading to inconsistent performance; 2) time-consuming and costly LLM querying. To address these challenges, we present MicLog, the first progressive meta in-context learning (ProgMeta-ICL) log parsing framework that combines meta-learning with ICL on small open-source LLMs (i.e., Qwen-2.5-3B). Specifically, MicLog: i) enhances LLMs' ICL capability through a zero-shot to k-shot ProgMeta-ICL paradigm, employing weighted DBSCAN candidate sampling and enhanced BM25 demonstration selection; ii) accelerates parsing via a multi-level pre-query cache that dynamically matches and refines recently parsed templates. Evaluated on Loghub-2.0, MicLog achieves 10.3% higher parsing accuracy than the state-of-the-art parser while reducing parsing time by 42.4%.
Open AccessMicLog: Towards Accurate and Efficient LLM-based Log Parsing via Progressive Meta In-Context Learning
Log parsing converts semi-structured logs into structured templates, forming a critical foundation for downstream analysis. Traditional syntax and semantic-based parsers often struggle with semantic variations in evolving logs and data scarcity stemming from their limited domain coverage. Recent large language model (LLM)-based parsers leverage in-context learning (ICL) to extract semantics from examples, demonstrating superior accuracy. However, LLM-based parsers face two main challenges: 1) underutilization of ICL capabilities, particularly in dynamic example selection and cross-domain generalization, leading to inconsistent performance; 2) time-consuming and costly LLM querying. To address these challenges, we present MicLog, the first progressive meta in-context learning (ProgMeta-ICL) log parsing framework that combines meta-learning with ICL on small open-source LLMs (i.e., Qwen-2.5-3B). Specifically, MicLog: i) enhances LLMs' ICL capability through a zero-shot to k-shot ProgMeta-ICL paradigm, employing weighted DBSCAN candidate sampling and enhanced BM25 demonstration selection; ii) accelerates parsing via a multi-level pre-query cache that dynamically matches and refines recently parsed templates. Evaluated on Loghub-2.0, MicLog achieves 10.3% higher parsing accuracy than the state-of-the-art parser while reducing parsing time by 42.4%.
- Open Access
Scalable Supervising Software Agents with Patch Reasoner
While large language model agents have advanced software engineering tasks, the unscalable nature of existing test-based supervision is limiting the potential improvement of data scaling. The reason is twofold: (1) building and running test sandbox is rather heavy and fragile, and (2) data with high-coverage tests is naturally rare and threatened by test hacking via edge cases. In this paper, we propose R4P, a patch verifier model to provide scalable rewards for training and testing SWE agents via reasoning. We consider that patch verification is fundamentally a reasoning task, mirroring how human repository maintainers review patches without writing and running new reproduction tests. To obtain sufficient reference and reduce the risk of reward hacking, R4P uses a group-wise objective for RL training, enabling it to verify multiple patches against each other's modification and gain a dense reward for stable training. R4P achieves 72.2% Acc. for verifying patches from SWE-bench-verified, surpassing OpenAI o3. To demonstrate R4P's practicality, we design and train a lite scaffold, Mini-SE, with pure reinforcement learning where all rewards are derived from R4P. As a result, Mini-SE achieves 26.2% Pass@1 on SWE-bench-verified, showing a 10.0% improvement over the original Qwen3-32B. This can be further improved to 32.8% with R4P for test-time scaling. Furthermore, R4P verifies patches within a second, 50x faster than testing on average. The stable scaling curves of rewards and accuracy along with high efficiency reflect R4P's practicality.
Open AccessScalable Supervising Software Agents with Patch Reasoner
While large language model agents have advanced software engineering tasks, the unscalable nature of existing test-based supervision is limiting the potential improvement of data scaling. The reason is twofold: (1) building and running test sandbox is rather heavy and fragile, and (2) data with high-coverage tests is naturally rare and threatened by test hacking via edge cases. In this paper, we propose R4P, a patch verifier model to provide scalable rewards for training and testing SWE agents via reasoning. We consider that patch verification is fundamentally a reasoning task, mirroring how human repository maintainers review patches without writing and running new reproduction tests. To obtain sufficient reference and reduce the risk of reward hacking, R4P uses a group-wise objective for RL training, enabling it to verify multiple patches against each other's modification and gain a dense reward for stable training. R4P achieves 72.2% Acc. for verifying patches from SWE-bench-verified, surpassing OpenAI o3. To demonstrate R4P's practicality, we design and train a lite scaffold, Mini-SE, with pure reinforcement learning where all rewards are derived from R4P. As a result, Mini-SE achieves 26.2% Pass@1 on SWE-bench-verified, showing a 10.0% improvement over the original Qwen3-32B. This can be further improved to 32.8% with R4P for test-time scaling. Furthermore, R4P verifies patches within a second, 50x faster than testing on average. The stable scaling curves of rewards and accuracy along with high efficiency reflect R4P's practicality.
- Open Access
CLEANet: Robust and Efficient Anomaly Detection in Contaminated Multivariate Time Series
Multivariate time series (MTS) anomaly detection is essential for maintaining the reliability of industrial systems, yet real-world deployment is hindered by two critical challenges: training data contamination (noises and hidden anomalies) and inefficient model inference. Existing unsupervised methods assume clean training data, but contamination distorts learned patterns and degrades detection accuracy. Meanwhile, complex deep models often overfit to contamination and suffer from high latency, limiting practical use. To address these challenges, we propose CLEANet, a robust and efficient anomaly detection framework in contaminated multivariate time series. CLEANet introduces a Contamination-Resilient Training Framework (CRTF) that mitigates the impact of corrupted samples through an adaptive reconstruction weighting strategy combined with clustering-guided contrastive learning, thereby enhancing robustness. To further avoid overfitting on contaminated data and improve computational efficiency, we design a lightweight conjugate MLP that disentangles temporal and cross-feature dependencies. Across five public datasets, CLEANet achieves up to 73.04% higher F1 and 81.28% lower runtime compared with ten state-of-the-art baselines. Furthermore, integrating CRTF into three advanced models yields an average 5.35% F1 gain, confirming its strong generalizability.
Open AccessCLEANet: Robust and Efficient Anomaly Detection in Contaminated Multivariate Time Series
Multivariate time series (MTS) anomaly detection is essential for maintaining the reliability of industrial systems, yet real-world deployment is hindered by two critical challenges: training data contamination (noises and hidden anomalies) and inefficient model inference. Existing unsupervised methods assume clean training data, but contamination distorts learned patterns and degrades detection accuracy. Meanwhile, complex deep models often overfit to contamination and suffer from high latency, limiting practical use. To address these challenges, we propose CLEANet, a robust and efficient anomaly detection framework in contaminated multivariate time series. CLEANet introduces a Contamination-Resilient Training Framework (CRTF) that mitigates the impact of corrupted samples through an adaptive reconstruction weighting strategy combined with clustering-guided contrastive learning, thereby enhancing robustness. To further avoid overfitting on contaminated data and improve computational efficiency, we design a lightweight conjugate MLP that disentangles temporal and cross-feature dependencies. Across five public datasets, CLEANet achieves up to 73.04% higher F1 and 81.28% lower runtime compared with ten state-of-the-art baselines. Furthermore, integrating CRTF into three advanced models yields an average 5.35% F1 gain, confirming its strong generalizability.
- Open Access
DynaCausal: Dynamic Causality-Aware Root Cause Analysis for Distributed Microservices
Cloud-native microservices enable rapid iteration and scalable deployment but also create complex, fast-evolving dependencies that challenge reliable diagnosis. Existing root cause analysis (RCA) approaches, even with multi-modal fusion of logs, traces, and metrics, remain limited in capturing dynamic behaviors and shifting service relationships. Three critical challenges persist: (i) inadequate modeling of cascading fault propagation, (ii) vulnerability to noise interference and concept drift in normal service behavior, and (iii) over-reliance on service deviation intensity that obscures true root causes. To address these challenges, we propose DynaCausal, a dynamic causality-aware framework for RCA in distributed microservice systems. DynaCausal unifies multi-modal dynamic signals to capture time-varying spatio-temporal dependencies through interaction-aware representation learning. It further introduces a dynamic contrastive mechanism to disentangle true fault indicators from contextual noise and adopts a causal-prioritized pairwise ranking objective to explicitly optimize causal attribution. Comprehensive evaluations on public benchmarks demonstrate that DynaCausal consistently surpasses state-of-the-art methods, attaining an average AC@1 of 0.63 with absolute gains from 0.25 to 0.46, and delivering both accurate and interpretable diagnoses in highly dynamic microservice environments.
Open AccessDynaCausal: Dynamic Causality-Aware Root Cause Analysis for Distributed Microservices
Cloud-native microservices enable rapid iteration and scalable deployment but also create complex, fast-evolving dependencies that challenge reliable diagnosis. Existing root cause analysis (RCA) approaches, even with multi-modal fusion of logs, traces, and metrics, remain limited in capturing dynamic behaviors and shifting service relationships. Three critical challenges persist: (i) inadequate modeling of cascading fault propagation, (ii) vulnerability to noise interference and concept drift in normal service behavior, and (iii) over-reliance on service deviation intensity that obscures true root causes. To address these challenges, we propose DynaCausal, a dynamic causality-aware framework for RCA in distributed microservice systems. DynaCausal unifies multi-modal dynamic signals to capture time-varying spatio-temporal dependencies through interaction-aware representation learning. It further introduces a dynamic contrastive mechanism to disentangle true fault indicators from contextual noise and adopts a causal-prioritized pairwise ranking objective to explicitly optimize causal attribution. Comprehensive evaluations on public benchmarks demonstrate that DynaCausal consistently surpasses state-of-the-art methods, attaining an average AC@1 of 0.63 with absolute gains from 0.25 to 0.46, and delivering both accurate and interpretable diagnoses in highly dynamic microservice environments.
- Open Access
Dual-Actor Fine-Tuning of VLA Models: A Talk-and-Tweak Human-in-the-Loop Approach
Vision-language-action (VLA) models demonstrate strong generalization in robotic manipulation but face challenges in complex, real-world tasks. While supervised fine-tuning with demonstrations is constrained by data quality, reinforcement learning (RL) offers a promising alternative. We propose a human-in-the-loop dual-actor fine-tuning framework grounded in RL. The framework integrates a primary actor for robust multi-task performance with a refinement actor for latent-space adaptation. Beyond standard physical interventions, we introduce a lightweight talk-and-tweak scheme that converts human corrections into semantically grounded language commands, thereby generating a new dataset for policy learning. In real-world multi-task experiments, our approach achieves 100% success across three tasks within 101 minutes of online fine-tuning. For long-horizon tasks, it sustains a 50% success rate over 12 consecutive operations. Furthermore, the framework scales effectively to multi-robot training, achieving up to a 2 times improvement in efficiency when using dual robots. The experiment videos are available at https://sites.google.com/view/hil-daft/.
Open AccessDual-Actor Fine-Tuning of VLA Models: A Talk-and-Tweak Human-in-the-Loop Approach
Vision-language-action (VLA) models demonstrate strong generalization in robotic manipulation but face challenges in complex, real-world tasks. While supervised fine-tuning with demonstrations is constrained by data quality, reinforcement learning (RL) offers a promising alternative. We propose a human-in-the-loop dual-actor fine-tuning framework grounded in RL. The framework integrates a primary actor for robust multi-task performance with a refinement actor for latent-space adaptation. Beyond standard physical interventions, we introduce a lightweight talk-and-tweak scheme that converts human corrections into semantically grounded language commands, thereby generating a new dataset for policy learning. In real-world multi-task experiments, our approach achieves 100% success across three tasks within 101 minutes of online fine-tuning. For long-horizon tasks, it sustains a 50% success rate over 12 consecutive operations. Furthermore, the framework scales effectively to multi-robot training, achieving up to a 2 times improvement in efficiency when using dual robots. The experiment videos are available at https://sites.google.com/view/hil-daft/.
- Open Access
Trust, But Verify: A Self-Verification Approach to Reinforcement Learning with Verifiable Rewards
Large Language Models (LLMs) show great promise in complex reasoning, with Reinforcement Learning with Verifiable Rewards (RLVR) being a key enhancement strategy. However, a prevalent issue is ``superficial self-reflection'', where models fail to robustly verify their own outputs. We introduce RISE (Reinforcing Reasoning with Self-Verification), a novel online RL framework designed to tackle this. RISE explicitly and simultaneously trains an LLM to improve both its problem-solving and self-verification abilities within a single, integrated RL process. The core mechanism involves leveraging verifiable rewards from an outcome verifier to provide on-the-fly feedback for both solution generation and self-verification tasks. In each iteration, the model generates solutions, then critiques its own on-policy generated solutions, with both trajectories contributing to the policy update. Extensive experiments on diverse mathematical reasoning benchmarks show that RISE consistently improves model's problem-solving accuracy while concurrently fostering strong self-verification skills. Our analyses highlight the advantages of online verification and the benefits of increased verification compute. Additionally, RISE models exhibit more frequent and accurate self-verification behaviors during reasoning. These advantages reinforce RISE as a flexible and effective path towards developing more robust and self-aware reasoners.
Open AccessTrust, But Verify: A Self-Verification Approach to Reinforcement Learning with Verifiable Rewards
Large Language Models (LLMs) show great promise in complex reasoning, with Reinforcement Learning with Verifiable Rewards (RLVR) being a key enhancement strategy. However, a prevalent issue is ``superficial self-reflection'', where models fail to robustly verify their own outputs. We introduce RISE (Reinforcing Reasoning with Self-Verification), a novel online RL framework designed to tackle this. RISE explicitly and simultaneously trains an LLM to improve both its problem-solving and self-verification abilities within a single, integrated RL process. The core mechanism involves leveraging verifiable rewards from an outcome verifier to provide on-the-fly feedback for both solution generation and self-verification tasks. In each iteration, the model generates solutions, then critiques its own on-policy generated solutions, with both trajectories contributing to the policy update. Extensive experiments on diverse mathematical reasoning benchmarks show that RISE consistently improves model's problem-solving accuracy while concurrently fostering strong self-verification skills. Our analyses highlight the advantages of online verification and the benefits of increased verification compute. Additionally, RISE models exhibit more frequent and accurate self-verification behaviors during reasoning. These advantages reinforce RISE as a flexible and effective path towards developing more robust and self-aware reasoners.
- Top Cited PapersOpen Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 3149-3167
Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training
Youliang Yuan, Wenxiang Jiao, Wenxuan Wang, Jen-tse Huang, Jiahao Xu, Tian Liang, Pinjia He, Zhaopeng Tu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 3149-3167
Top Cited PapersOpen AccessRefuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training
Youliang Yuan, Wenxiang Jiao, Wenxuan Wang, Jen-tse Huang, Jiahao Xu, Tian Liang, Pinjia He, Zhaopeng Tu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 17825-17846
Insight Over Sight: Exploring the Vision-Knowledge Conflicts in Multimodal LLMs
Xiaoyuan Liu, Wenxuan Wang, Youliang Yuan, Jen-tse Huang, Qiuzhi Liu, Pinjia He, Zhaopeng Tu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 17825-17846
Open AccessInsight Over Sight: Exploring the Vision-Knowledge Conflicts in Multimodal LLMs
Xiaoyuan Liu, Wenxuan Wang, Youliang Yuan, Jen-tse Huang, Qiuzhi Liu, Pinjia He, Zhaopeng Tu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.
- Open Access
Difficult Task Yes but Simple Task No: Unveiling the Laziness in Multimodal LLMs
Multimodal Large Language Models (MLLMs) demonstrate a strong understanding of the real world and can even handle complex tasks. However, they still fail on some straightforward visual question-answering (VQA) problems. This paper dives deeper into this issue, revealing that models tend to err when answering easy questions (e.g. Yes/No questions) about an image, even though they can correctly describe it. We refer to this model behavior discrepancy between difficult and simple questions as model laziness. To systematically investigate model laziness, we manually construct LazyBench, a benchmark that includes Yes/No, multiple choice, short answer questions, and image description tasks that are related to the same subjects in the images. Based on LazyBench, we observe that laziness widely exists in current advanced MLLMs (e.g. GPT-4o, Gemini-1.5-pro, Claude 3 and LLaVA-v1.5-13B), and it is more pronounced on stronger models. We also analyze the VQA v2 (LLaVA-v1.5-13B) benchmark and find that about half of its failure cases are caused by model laziness, which further highlights the importance of ensuring that the model fully utilizes its capability. To this end, we conduct preliminary exploration on how to mitigate laziness and find that chain of thought (CoT) can effectively address this issue.
Open AccessDifficult Task Yes but Simple Task No: Unveiling the Laziness in Multimodal LLMs
Multimodal Large Language Models (MLLMs) demonstrate a strong understanding of the real world and can even handle complex tasks. However, they still fail on some straightforward visual question-answering (VQA) problems. This paper dives deeper into this issue, revealing that models tend to err when answering easy questions (e.g. Yes/No questions) about an image, even though they can correctly describe it. We refer to this model behavior discrepancy between difficult and simple questions as model laziness. To systematically investigate model laziness, we manually construct LazyBench, a benchmark that includes Yes/No, multiple choice, short answer questions, and image description tasks that are related to the same subjects in the images. Based on LazyBench, we observe that laziness widely exists in current advanced MLLMs (e.g. GPT-4o, Gemini-1.5-pro, Claude 3 and LLaVA-v1.5-13B), and it is more pronounced on stronger models. We also analyze the VQA v2 (LLaVA-v1.5-13B) benchmark and find that about half of its failure cases are caused by model laziness, which further highlights the importance of ensuring that the model fully utilizes its capability. To this end, we conduct preliminary exploration on how to mitigate laziness and find that chain of thought (CoT) can effectively address this issue.