Lunch at 12:30pm, talk at 1pm, in 148 Fitzpatrick

Title: Function Call Graph Encoding for Neural Source Code Summarization

Abstract: Automated source code summarization has been a long standing goal for the software engineering community. The current research aims at generating summaries for short snippets of code, usually a function. Since the emergence of Neural Networks and NMT models, researchers have leveraged states of the art NLP techniques. However, in contrast to natural language datasets, the quality of data available for code summarization is missing. Programmers often write poor quality natural language comments. More importantly, every programmer has their own shorthand for describing what their snippet of code does. My research has been aimed at bringing more context beyond the code-description pair to improve the quality of these summaries.

In this talk, I will present one of my latest projects that encodes several functions from the program curated via a function call graph. The aim of paper was to explore the possibility of helpful information outside the code-description pair. This function call graph, with the help of attention mechanism adds context-awareness to the model to improve the quality of summaries.

Bio: Aakash is a Ph.D. student at the university of Notre Dame advised by Prof. Collin McMillan. His current research leverages the intersection of NLP and Software Engineering for code summarization and program comprehension. His broad research interests are aimed at working towards intelligent systems.