Matthew W. Hoffman
I am currently a graduate student in the CS department at UBC,
supervised by Nando de Freitas and Arnaud Doucet. I’m working on
various approaches to reformulate stochastic control problems as
inference problems, particularly in large, continuous domains.
You can also take a look at my CV, although this
may be horribly out of date.
KAIST ML Tutorial
In a recent tutorial for KAIST entitled “Reinforcement learning and
planning”, I made a few quick demos of dynamic programming and
reinforcement learning methods, which can be found here. My slides for this presentation can
be found here.
In the run-up to my soon-to-be-coming graduation I’ve been
collecting all of my code and making that available, all of which
you can find on the Software page.
- M. Hoffman, A. Lazaric, M. Ghavamzadeh, R. Munos.
Regularized Least Squares Temporal Difference learning with
nested l2 and l1 penalization. EWRL,
- M. Hoffman, E. Brochu, N. de Freitas. Portfolio Allocation
for Bayesian Optimization. UAI, 2011.
- M. Ghavamzadeh, A. Lazaric, R. Munos, M. Hoffman. Finite
sample analysis of Lasso-TD. ICML,
- M. Hoffman and N. de Freitas. Inference strategies for
solving semi-Markov decision processes. To
appear in Decision Theory Models for Applications in Artificial
Intelligence: Concepts and Solutions, L.E. Sucar, E. Morales, H. Hoey
- M. Hoffman, P. Carbonetto, N. de Freitas, and A. Doucet.
Inference strategies for solving semi-Markov decision
processes. NIPS Workshop on Probabilistic
Approaches for Robotics and Control, 2009.
- M. Hoffman, H. Kueck, N. de Freitas, A. Doucet. New
inference strategies for solving Markov Decision Processes using
reversible jump MCMC. UAI, 2009.
- H. Kueck, M. Hoffman, A. Doucet, N. de Freitas. Inference
and Learning for Active Sensing, Experimental Design, and Control. Invited paper, IBPRIA, 2009.
- M. Hoffman, N. de Freitas, A. Doucet, Jan Peters. An
Expectation Maximization Algorithm for Continuous Markov Decision
Processes with Arbitrary Rewards. AISTATS,
- M. Hoffman, A. Doucet, N. de Freitas, and A. Jasra.
Bayesian policy learning with trans-dimensional MCMC. NIPS, 2007. (This is an extended and
greatly revised version of UBC CS TR-2007-04.)
- M. Hoffman, A. Doucet, N. de Freitas, and A. Jasra. On
Solving General State-Space Sequential Decision Problems using
Inference Algorithms. UBC CS
TR-2007-04. March, 2007.
In a past life I was an undergraduate in the CS and Math departments
at the University of Washington. While there I worked with Rajesh
Rao as part of the Neural Systems Group, and I focused primarily
on problems of gaze-imitation and imitation-learning utilizing
shared-attention. For further information see the relevant publications: