Open access
2026
TokenGated-CodeBERT With Lightweight Attention-Based Token Selection for Software Defect Prediction
TokenGated-CodeBERT is proposed, a new methodology designed for scenarios with severe class imbalance that supports full end-to-end training, keeps the original CodeBERT encoder intact, requires no additional auxiliary optimization objectives, and features a lightweight architecture.
Bin Shuai, De-Quan Xu, Yuan-Lin Yang
· IEEE Access · 0 citations