"instance-dependent analysis of learning algorithms"
Abstract: Classical statistical learning theory uses probabilistic models to analyze the trade-off between the amount of data and the achievable accuracy of learning for a given problem. However, the resulting theory is both brittle (what if the assumptions are not met?) and can also be pessimistic as a result of theorists often focusing on obtaining worst-case guarantees. Contrary to this, practitioners find that learning algorithms often work outside of the scope of theoretical results. Can we refine learning theory to bridge this gap? In this talk an attempt will be made to demonstrate that this is possible by performing instance-dependent analysis in three quite different learning settings. One of the settings chosen is an unsupervised problem (independent component analysis), the second one is a least square problem, and the last one is an online linear prediction problem. Results will be given that explain in detail various earlier observed gaps between theory and practice. The broader claim is that with the approach presented, learning theory can successfully contribute beyond its usual boundaries, by providing refined performance bounds with little or no assumptions on data generation and eventually leading to better algorithms.
ai seminar series
Fridays at noon, Amii and the Department of Computing Science host AI Seminars, engaging presentations on topics in the broad field of artificial intelligence. With speakers from the University of Alberta and other world-leading groups, the talks give AI enthusiasts a friendly way of engaging with the latest trends and topics in research and development.
Seminars are open to the public, and no registration is required, though seating is limited and on a first-come-first-served basis. Topics range from foundational theoretical work to innovative applications of artificial intelligence technologies.
If you would like to present at an upcoming AI Seminar, please contact Colin Bellinger.
Join the AI Seminar mailing list to stay up-to-date on all the latest presentations.