“Limits of Transfer Learning,” a paper by the Harvey Mudd College Walter Bradley Center Clinic team, has been accepted to the Sixth International Conference on Machine Learning, Optimization and Data Science (LOD 2020).
The Computer Science Clinic team—advisor and computer science professor George Montañez, Jake Williams ’20, Tyler Sam ’20, Abel Tadesse ’20 (CMC) and Huey Sun ’20 (POM)—explores transfer learning, a current frontier of machine learning in which insight gained from solving one problem is applied to solve a separate, but related problem.
“The paper places theoretical limits on transfer learning,” says Montañez. “Not much theoretical work has been done for transfer learning, even though it is widely used in practice. Our paper fills an existing gap in knowledge by addressing what can lead to successful transfer learning and what limitations exist for its widespread adoption.”
Members of the team will present the paper at LOD 2020 via Zoom later this month. “The past two conferences have had acceptance rates of approximately 35%, so it is a fairly competitive conference to get into,” says Montañez.
This is the fifth paper Montañez has published with students this year and the second paper from this Clinic project accepted to a major conference. The first (Decomposable Probability-of-Success Metrics in Algorithmic Search) was presented at the 12th International Conference on Agents and Artificial Intelligence in February.