Practical and Scalable Inference for Deep Gaussian Processes, Maurizio Fillippone, bayesgroup.ru

The study of complex phenomena through the analysis of data often requires us to make assumptions about the underlying dynamics.
In modern applications, for many systems of interest we are facing the challenge of doing so when very little is known about their mechanistic description.
Even when a mechanistic description is available, simulating such systems is so computationally expensive that we cannot use it effectively.
Probabilistic models based on Deep Gaussian Processes (DGPs) offer attractive tools to tackle these challenges in a principled way and to allow for a sound quantification of uncertainty.
However, inference for DGPs poses huge computational challenges that arguably hinder their wide adoption.
In this talk, I will present our contribution to the development of practical and scalable inference for DGPs, which can exploit distributed and GPU computing.
In particular, I will introduce a novel formulation of DGPs based on random features that we infer using stochastic variational inference.
Through a series of experiments, I will illustrate how our proposal enables scalable deep probabilistic nonparametric modeling and significantly advances the state-of-the-art on inference methods for DGPs.

Скачать видео