This talk has been canceled and will be rescheduled for Fall 2024.

When: Canceled

Where: Canceled

Speaker: Nicholas LaRacuente

Title: Learning Across Complex Systems to Use Small, Noisy Datasets Effectively

Abstract:

There is often a gap between how explainable a complex, non-equilibrated system appears and how well plausible models verifiably explain or predict it. This situation is common for ecological time series – repeated observations from the same system at different times. Many such time series are noisy, short, and full of illusory correlations and explanations. In light of the bias-variance tradeoff, adding features to a model often worsens predictions even if those features are mechanistically sensible. By leveraging outside knowledge, however, models can escape some tradeoffs of constrained training data. I will show preliminary results on strategies for transferring domain knowledge and other external information. Finally, I note some possible implications in quantum machine learning and simulation.

This talk includes preliminary joint work with Mohsen Heidari Khoozani and Alireza Tavakoli.

Biography: Nicholas LaRacuente is an assistant professor of computer science at UI Bloomington since autumn 2023. Previously, he obtained a PhD in physics and concurrent master’s in mathematics at the University of Illinois at Urbana-Champaign, then held a postdoctoral appointment at the University of Chicago. Nicholas’s research interests include computational complexity, quantum computing, and ways in which computation and information manifest in nature.