Radford Neal's Research Interests

I view statistics as the general science of empirical learning, which can be used to support scientific inference, to solve engineering problems, and to clarify how natural and artificial agents can learn from experience.

Statistics is not always easy. It raises many interesting philosophical issues, and often requires the solution of difficult computational problems. Here are some problems and solutions of particular interest to me (click on title for more details):

Bayesian inference

An approach to statistics in which all forms of uncertainty are expressed in terms of probability.
Markov chain Monte Carlo
A way of computing high-dimensional integrals that is crucial for doing Bayesian inference.
Neural networks
Statistical models that are relevant to, or at least inspired by, the way learning and computation may occur in the brain.
Latent variable models
Models phrased in terms of entities that we have invented to explain patterns we see in observable variables.
Evaluation of learning methods
Ways of telling which methods for learning from data really work.
Data compression
Using models for data to find a compressed representation of it.
Error correcting codes
Representing information in a redundant form that allows errors to be corrected with high probability.
Statistical applications
I have worked on various statistical applications, mostly of a biological nature.
I also have current, dormant, or possible future interests in philosophy of science, artificial life, programming languages, user interface design, and who knows what else...


Back to Radford Neal's home page