Learning under model misspecification

Add to your list(s) Download to your calendar using vCal

• Andrés R. Masegosa, Universidad de Almería (Spain)
• Tuesday 02 February 2021, 13:15-14:15
• Zoom.

If you have a question about this talk, please contact Mateja Jamnik.

Join us on Zoom

Bayesian statistics is one of the most employed tools for uncertainty modelling in deep learning. But, in this talk, we will present recent research which cast doubts about the optimality of the Bayesian approach for this task. More precisely, we will present a novel PAC -Bayesian analysis of Bayesian model averaging showing that it is only optimal for generalization when the model class is perfectly specified, something which rarely happens in practice.

By building on this theoretical analysis, we will introduce a novel learning framework based on the minimization of a new family of PAC -Bayesian bounds which explicitly assume that the model class is misspecified (a much more realistic assumption). We will also discuss strong connections with deep ensemble methods.

This talk is part of the Artificial Intelligence Research Group Talks (Computer Laboratory) series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• Artificial Intelligence Research Group Talks (Computer Laboratory)
• Cambridge Centre for Data-Driven Discovery (C2D3)
• Cambridge Forum of Science and Humanities
• Cambridge Language Sciences
• Cambridge talks
• Chris Davis' list
• Department of Computer Science and Technology talks and seminars
• Guy Emerson's list
• Hanchen DaDaDash
• Interested Talks
• School of Technology
• Speech Seminars
• Trust & Technology Initiative - interesting events
• Zoom
• bld31
• ndk22's list
• ob366-ai4er
• rp587
• yk373's list
• yk449

Note that ex-directory lists are not shown.