AI for Software Maintenance
Principal Investigator: Sandareka Wickramanayake
We propose to advance automated program repair by leveraging natural language processing techniques, iterative repair strategies, and explainable AI frameworks to enhance accuracy, robustness, and interpretability of repair systems.
Automated program repair (APR) seeks to reduce the effort required to identify and fix errors in source code by developing systems capable of suggesting or applying corrections automatically. While significant progress has been made, challenges remain in handling complex, multi-line bugs and in ensuring that repair systems are both accurate and interpretable.
This project investigates how natural language processing methods can be applied to model source code for automated repair tasks. Iterative and multi-line repair strategies are explored to better capture the context and dependencies within code, improving the quality of generated fixes. Fusion-based approaches are designed to integrate multiple repair techniques, enabling complementary strengths to be leveraged within a single framework.
In addition, explainable AI (XAI) methods are incorporated to improve the transparency and trustworthiness of repair recommendations. Evaluation frameworks and benchmarks are proposed to systematically measure the effectiveness, robustness, and interpretability of APR systems. By combining NLP-driven modeling, iterative strategies, and explainability, this research contributes toward the development of more powerful and reliable automated program repair systems.
Objectives:
- Develop natural language processing methods to model and understand source code for automated repair tasks.
- Investigate iterative and multi-line repair strategies that improve the accuracy and robustness of automated fixes.
- Explore explainable AI (XAI) frameworks to enhance the interpretability and transparency of repair suggestions.
- Integrate fusion-based and hybrid approaches to combine multiple repair strategies effectively.
- Establish evaluation methods and benchmarks to assess performance, interpretability, and generalizability of automated program repair systems.
Keywords: Machine Learning / Deep Learning | Natural Language Processing | Automated Program Repair | Code Generation |
Publications
Workshop Papers
Jayanath Senevirathna, Ayesh Vininda, Prasad Sandaruwan, Ridwan Shariffdeen, Sandareka Wickramanayake, and Nisansa de Silva, "FusionRepair: Iterative Multi-Line APR via Fusion", in IEEE/ACM International Workshop on Automated Program Repair (APR), 2025, pp. 27--34. doi: 10.1109/APR66717.2025.00009
Nethum Lamahewage, Nimantha Cooray, Ridwan Shariffdeen, Sandareka Wickramanayake, and Nisansa de Silva, "SCHOLIA-An XAI Framework for APR", in IEEE/ACM International Workshop on Automated Program Repair (APR), 2025, pp. 19--26. doi: 10.1109/APR66717.2025.00008
Extended Abstracts
H L S Sankaja, R P G Sandeeptha, R M R U Ranasingha, S Wickramanayake, and Nisansa de Silva, "Generating Natural Language Explanations for Automated Program Repair Generated Patches", in International Conference of Sabaragamuwa University of Sri Lanka, 2025, pp. 61.
Preprints
Kaushal Kahapola, Sharada Galappaththi, Dinith Ranasinghe, Ridwan Shariffdeen, Nisansa de Silva, Srinath Perera, and Sandareka Wickramanayake, "JavaBackports: A Dataset for Benchmarking Automated Backporting in Java", in 23rd International Conference on Mining Software Repositories (MSR), 2026.
Mineth Weerasinghe, Himindu Kularathne, Methmini Madhushika, Danuka Lakshan, Nisansa de Silva, Adeesha Wijayasiri, and Srinath Perera, "From Monolith to Microservices: A Comparative Evaluation of Decomposition Frameworks", arXiv preprint arXiv:2601.23141, 2026
Team
External Collaborators: | Ridwan Shariffdeen | Srinath Perera | Adeesha Wijayasiri |
