Towards improving automated unit test generation for machine learning libraries using structured input data

Abstract

This talk is related to my Master’s thesis and my research internship, which are about to end. I will present some of the challenges that frequently arise when using search-based automated unit test generation tools on machine learning libraries in Python. Then, I will talk about the solution I have implemented to solve these challenges based on a tool named Pynguin and on grammar-based fuzzing techniques. Finally, I’ll compare the performance of my solution with the performance of the original tool.