 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
|||||||||||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
|||
 |
 |
 |
 |
|
||||||||||
 |
|
|||||||||||||
 |
|
|||||||||||||
 |
 |
|
||||||||||||
| Seminar Schedule (Seminar Organizer: Prof. Ziv Bar-Joseph) | |
Machine Learning Lunch Seminar
|
Spring 2009 Seminar
Date: April 8, 2009 Time: 10:00 AM - 11:30 PM Location: 1507 Newell-Simon Hall Speaker: Ryan Adams Ph.D. Student Title: Tractable Nonparametric Bayesian Inference in the Poisson Process Abstract: The Poisson process is a ubiquitous model for time series and spatial data. The inhomogeneous variant of the Poisson process allows the rate of arrivals to vary in time (or space). When performing Bayesian inference of a Poisson rate function, we would prefer to make as few assumptions as possible by using a nonparametric prior. The Gaussian process is a particularly appealing prior for the rate function, resulting in the Gaussian Cox process. Unfortunately, fully-nonparametric Bayesian inference in the Gaussian Cox process has been impossible, due to the need to integrate an infinite-dimensional random function. In this talk, I will describe how we have developed an alternative Gaussian Cox process construction that allows tractable inference using recently-developed Markov chain Monte Carlo methods. This model is the first to enable fully-nonparametric kernel-based Bayesian inference in the inhomogeneous Poisson process without requiring a crippled model or a finite-dimensional approximation. This is joint work with Iain Murray (University of Toronto) and David J.C. MacKay (University of Cambridge).
