Neil Lawrence: New Perspectives on Variational Approximations in Gaussian Processes: Modelling Data

This talk was written in the train on the way down to Cambridge from Sheffield: I was there to examine Andrew Wilson's PhD thesis. It's a really interesting group to speak to, and Thang Bui, who was at the GP Winter School the week before asked if I could give a talk. I initially said no, because I felt I didn't have anything new to say that was prepared, but on reflection I thought it would be a fun idea to put something together on the train and do a whiteboard talk (for the first time in a long time). The nice thing is that most of the preparation was done walking around thinking about stuff, and it was a chance to collect a few strands of my thinking, to try and achieve a coherent result. This was the result, whether it worked or not the viewer can judge. The abstract I provided was intended to keep expectations low (although I think not everyone read it!).

I was originally just going to do an audio recording for my own reference, but Richard Turner kindly sorted out the video. Unfortunately the microphone I'm wearing was turned off. Apologies for the poor sound quality. Thanks to David Lopez-Paz, Patrick Gosling and Carl Rasmussen for sorting it out and getting me the video.

Abstract

In this talk I'll introduce new perspectives on variational approximations. Many of the ideas may be widely applicable, but we will try to instantiate them in the context of Gaussian process models.

Although the variational material itself is reasonably technical, I'll try and start the talk by making general statements about data modelling. Then, in an effort to make the talk seem coherent, I'll make claims that the technical material which follows was inspired by the wider perspective I've given. Of course in practice, the technical material really emerged across a number of years during discussions with many people, and the general perspective has been retrofitted. Still, I'll be giving the talk amongst friends, so no one will mind too much if the story doesn't really fit together, and in fact it might be a good trigger for discussion. Speaking of which, I'll be looking forward to lots of audience participation, and such participation may take the talk in previously unplanned directions.

The talk will be given without the use of electronic aids.