The Patch Overfitting Problem in Automated Program Repair: Practical Magnitude and a Baseline for Realistic Benchmarking

Published: July 10, 2024

Language: Английский

---

JIT-Smart: A Multi-task Learning Framework for Just-in-Time Defect Prediction and Localization

Proceedings of the ACM on software engineering., Journal Year: 2024, Volume and Issue: 1(FSE), P. 1 - 23

Published: July 12, 2024

Just-in-time defect prediction (JIT-DP) is used to predict the defect-proneness of a commit and just-in-time localization (JIT-DL) locate exact buggy positions (defective lines) in commit. Recently, various JIT-DP JIT-DL techniques have been proposed, while most them use post-mortem way (e.g., code entropy, attention weight, LIME) achieve goal based on results JIT-DP. These methods do not utilize label information defective lines during model building. In this paper, we propose unified JIT-Smart, which makes training process tasks mutually reinforcing multi-task learning process. Specifically, design novel network (DLN), explicitly introduces for supervised with considering class imbalance issue. To further investigate accuracy cost-effectiveness compare JIT-Smart 7 state-of-the-art baselines under 5 commit-level line-level evaluation metrics JIT-DL. The demonstrate that statistically better than all JIT-DP, at median value, achieves F1-Score 0.475, AUC 0.886, Recall@20%Effort 0.823, Effort@20%Recall 0.01 Popt 0.942 improves by 19.89%-702.74%, 1.23%-31.34%, 9.44%-33.16%, 21.6%-53.82% 1.94%-34.89%, respectively . JIT-DL, Top-5 Accuracy 0.539 Top-10 0.396, line 0.726, 0.087 IFA 0.098 101.83%-178.35%, 101.01%-277.31%, 257.88%-404.63%, 71.91%-74.31% 99.11%-99.41%, respectively. Statistical analysis shows our performs more stably best-performing model. Besides, also best performance compared cross-project evaluation.

Language: Английский

---

Enhancing the Efficiency of Automated Program Repair via Greybox Analysis

Published: Oct. 18, 2024

Language: Английский

---

Shibboleth: Hybrid Patch Correctness Assessment in Automated Program Repair

Published: Oct. 10, 2022

Test-based generate-and-validate automated program repair (APR) systems generate many patches that pass the test suite without fixing bug. The generated must be manually inspected by developers, a task tends to time-consuming, thereby diminishing role of APR in reducing debugging costs.

Language: Английский

---