2021-: Code Migration for ML-based Software (Danny Dig)

Project Abstract: It is widely known that at least two-thirds of software costs are due to evolution, with some industrial surveys claiming 90%. ML software and models need to evolve to respond to internal & external changes. An important example of evolution is to migrate code to use the newer version of ML libraries or to optimize performance. Due to many non-backwards compatible API changes, this often requires engineers to rewrite their code & models from scratch. Our goal is to mechanize such tasks that are expensive, time-consuming, and error prone. In this project we first automatically mine evolutionary code changes from a wide repository of open-source ML codebases. Grounded on these formative studies, we will design, implement, and evaluate refactoring tools to help engineers evolve their codebases safely and effectively.

TEAM

Faculty: Danny Dig (CU)

Students: Malinda Dilhara (CU - 3rd year PhD student), Ameya Ketkar (OSU - 5th year PhD student, will graduate in Fall 2021)