What’s New & Updated

Posts about papers, events, news, and more

New Appointment at SUTD

I am thrilled to share that I will be joining the ISTD Pillar, Singapore University of Technology and Design (SUTD) as an Assistant Professor, following the completion of my PhD early next year. I am deeply grateful to everyone who has supported me throughout this journey. Looking ahead, I am excited to start building my research group at SUTD. I am currently seeking motivated RAs and PhD students, and I am eager to explore new collaborations. If you are interested in working together, please feel free to reach out to me at thanhlc@ieee.org.
news
Dec 23, 2025
Thanh Le-Cong

Accepted Paper by ICSE 2026 - Second Round

I am excited to share that our work on ‘How Do Semantically Equivalent Code Transformations Impact Membership Inference on LLMs for Code?’ has been accepted by Research Track of ICSE 2026.
paper
news

The success of large language models for code relies on vast amounts of code data, including public open-source repositories, such as GitHub, and private, confidential code from companies. This raises concerns about intellectual property compliance and the potential unauthorized use of license-restricted code. While membership inference (MI) techniques have been proposed to detect such unauthorized usage, their effectiveness can be undermined by semantically equivalent code transformation techniques, which modify code syntax while preserving semantic. In this work, we systematically investigate whether semantically equivalent code transformation rules might be leveraged to evade MI detection. The results reveal that model accuracy drops by only 1.5% in the worst case for each rule, demonstrating that transformed datasets can effectively serve as substitutes for fine-tuning. Additionally, we find that one of the rules (RenameVariable) reduces MI success by 10.19%, highlighting its potential to obscure the presence of restricted code. To validate these findings, we conduct a causal analysis confirming that variable renaming has the strongest causal effect in disrupting MI detection. Notably, we find that combining multiple transformations does not further reduce MI effectiveness. Our results expose a critical loophole in license compliance enforcement for training large language models for code, showing that MI detection can be substantially weakened by transformation-based obfuscation techniques.

Dec 17, 2025
Hua Yang, Alejandro Velasco, Thanh Le-Cong, Md Nazmul Haque, Bowen Xu, Denys Poshyvanyk

PhD Completion Seminar

I’m thrilled to share that after three incredible years at The University of Melbourne, I have presented my PhD completion seminar! 🎓 This has been a journey of massive growth. Thank you to my supervisors, Dr. Bach Le and Prof. Toby Murray, for being not just mentors but true role models. And to my family and friends–thank you for keeping me grounded and celebrating the small wins. I’m really excited for what comes next!
talk
news
Nov 18, 2025
Thanh Le-Cong

Invited Talk at SUTD

Last week, I’m honored to be giving my talk ‘From Smart Tools to Trusted Teammates: Towards Trustworthy Automatic Programming’ at the Singapore University of Technology and Design (SUTD). Thank you to faculty and staff members of SUTD for this incredible opportunity and your generous hospitality. It’s a privilege to share my latest work and learn from yours during my visit.
talk
news
Nov 7, 2025
Thanh Le-Cong

Accepted Paper by ICSE 2026

I am excited to share that FLAMES has been accepted by Research Track of ICSE 2026. FLAMES is a memory-efficient LLM-based APR with semantic-guided patch generation. If you are interested in FLAMES, please read our pre-print and explore the tool.
paper
presentation
news

In this work, we unveil the inefficiency of LLM-based APR techniques, which leads to a significant decrease of their performance in constrained environments. To address this challenge, we propose FLAMES, an efficient LLM-based APR approach that overcomes the memory inefficiency of traditional beam search methods. Instead of beam search, FLAMES employs a semantic-guided best-first search with greedy decoding, using test feedback to guide patch generation. Experiments show that FLAMES reduces memory usage by up to 83% while achieving state-of-the-art repair accuracy, fixing 133 bugs on Defects4J and outperforming baselines by 12% on HumanEval-Java and 36.5% on TransformedD4J.

Oct 16, 2025
Thanh Le-Cong, Bach Le, Toby Murray

PC Members of FORGE 2026 - SVM 2026

I’m happy to be a program committee member of The 3rd ACM international conference on AI Foundation Models and Software Engineering (FORGE 2026) and International Workshop on Software Vulnerability Management (SVM 2026). Both events will be co-located with ICSE 2026. Please consider to submit your relevant works to these venues!
news
Sep 22, 2025
Thanh Le-Cong

Invited Talk at SoftwareLab@CISPA

This week, I’m honored to be giving my talk ‘Beyond Accuracy: A Closer Look on the Reliability of Large Language Models for Code’ at the Software Lab at the CISPA Helmholtz Center for Information Security. Thank you to Professor Michael Pradel and members of Software Lab for this incredible opportunity and your generous hospitality. It’s a privilege to share my latest work and learn from yours during my visit.
talk
news
Sep 2, 2025
Thanh Le-Cong

Invited Talk at CREST@UofA

This week, I’m honored to be giving my talk ‘Toward Reliable LLM-assisted Programming’ at the ’Centre for Research on Engineering Software Technologies (CREST) at the University of Adelaide. Thank you to Professor Ali Babar, Dr. Triet Le, and the entire CREST team for this incredible opportunity and your generous hospitality. It’s a privilege to share my latest work and learn from yours during my visit.
talk
news
Jun 23, 2025
Thanh Le-Cong

Accepted Paper by ACL 2025

I am excited to share that FormalBench has been accepted by Main Conference of ACL 2025 with an acceptance rate of 20.3%.
paper
presentation
news

In this work, we introduce a new reasoning task, i.e., specification inference, to evaluate how well Large Language Models (LLMs) can reason about program semantics. This work also includes a large dataset of C/Java programs and automated evaluation metrics to assess consistency and completeness of LLM-generated specification.

Jun 16, 2025
Thanh Le-Cong, Bach Le, Toby Murray

Accepted Paper by ICSME 2025

I am excited to share that our research on vulnerability prediction, led by my student - Duong Nguyen, has been (directly) accepted by Research Track of ICSME 2025 with an early acceptance rate of 6.4%. Congrats, Duong!
news

JIT-VP models are considered as potential solutions for proactive software security, promising to catch vulnerabilities the moment they’re committed with impressive detection rates of 80% and 90% precision. But are these published models truly ready for industry deployment? Our work reveals that their performance has been vastly overestimated due to testing in artificially balanced environments and drops significantly in a realistic scenario.

May 1, 2025

Accepted Paper by ACM TOSEM

I am excited to share that my first-authored paper on natural robustness of Code LLMs in Program Repair has been accepted by ACM Transaction on Software Engineering and Methodology.
news

Prior research often underestimates the robustness of Code LLMs due to reliance on unnatural code transformations in evaluation. To address this, we introduce human-validated criteria and a naturalness metric, demonstrating that many transformations lack real-world relevance and distort assessments of Code LLM effectiveness in program repair. Nevertheless, our re-examination of natural robustness shows that Code LLMs remain unrobust in program repair, with substantial performance declines on naturally occurring variants.

Mar 1, 2025

Quarto Academic Website Examples and Tips

news

Examples and tips for Quarto Academic Website

Oct 15, 2024
Dr. Gang He

News item

news

About new item

Jun 24, 2024
News
No matching items