Improving automated program repair using two-layer tree-based neural networks

Document Type

Conference Proceeding

Publication Date

6-27-2020

Abstract

We present DLFix, a two-layer tree-based model learning bug-fixingcode changes and their surrounding code context to improve Automated Program Repair (APR). The first layer learns the surroundingcode context of a fix and uses it as weights for the second layer thatis used to learn the bug-fixing code transformation. Our empiricalresults on Defect4J show that DLFix can fix 30 bugs and its resultsare comparable and complementary to the best performing patternbased APR tools. Furthermore, DLFix can fix 2.5 times more bugsthan the best performing deep learning baseline.

Identifier

85094111380 (Scopus)

ISBN

[9781450371223]

Publication Title

Proceedings International Conference on Software Engineering

External Full Text Location

https://doi.org/10.1145/3377812.3390896

ISSN

02705257

First Page

316

Last Page

317

Grant

CCF-1518897

Fund Ref

National Science Foundation

This document is currently not available here.

Share

COinS