Kamélia Daudel: Learning with importance weighted variational inference
Kamélia Daudel is an Assistant Professor in the Department Information Systems, Data Analytics and Operations at ESSEC Business School in France. Her current research is within the field of approximate inference. She is, in particular, interested in variational inference methods which go beyond the commonly used parametric variational distribution framework and which involve flexible families of divergence measures.
Time and place:–,
Erling Sverdrups plass, Niels Henrik Abels hus, 8th floor
This talk explores the methodological and computational challenges of using importance-weighted bounds for approximate Bayesian inference within the framework of variational inference. While standard variational inference typically relies on the Evidence Lower BOund (ELBO) to approximate posterior densities, moving to a broader family of importance-weighted variational inference bounds introduces gradient optimization difficulties. We address these difficulties through dedicated asymptotic analyses and the development of advanced variance-reduction strategies.
