In this talk, I will explain how fundamental problems in computational learning theory are at the heart of modern problems in machine learning and scientific applications and how algorithmic insights in mathematically tractable models can inspire new solutions in a wide variety of domains. I will explore two directions. First, I will explore algorithmic foundations for model stealing of language models.  Model stealing, where a learner tries to recover an unknown model through query access, is a critical problem in machine learning. Here, I will aim to build a theoretical foundation for designing model stealing algorithms.  Second, I will introduce Hamiltonian learning, a central computational task towards understanding and benchmarking quantum systems.  I will highlight how the lens of learning theory plays a key role in identifying and circumventing previous barriers and allows us to give efficient algorithms in settings that were previously conjectured to be intractable.

Speaker bio: Allen Liu is currently a fifth-year graduate student in EECS at MIT, advised by Ankur Moitra. His research is in learning theory, broadly defined, encompassing classical learning theory and statistics, as well as problems in modern machine learning and scientific applications such as quantum information.  His work has been awarded Best Student Paper at QIP in 2024 and featured in popular science media such as Quanta Magazine’s Biggest Breakthroughs in Computer Science for 2024.