Job talk for Assistant/Associate Professor in Computer Science
"From the sequence memoizer to the infinite, structured,
explicit-duration hidden Markov model: recent
developments in expressive, general-purpose,
computationally-efficient probabilistic modeling"
Frank Wood
Tuesday, April 24, 2012
11 a.m.
HR 10
ABSTRACT: Computational manipulation of uncertainty in the form of
probability is of growing importance to the sciences in general, is the
cornerstone of machine learning research, and is central to modern
approaches to artificial intelligence. All of the above drive the
development of increasingly expressive and efficient models. Some
migrate and become broadly applied in the sciences. This talk will
introduce two that are primed for such a migration: the sequence
memoizer (SM) and the infinite, structured, explicit-duration hidden
Markov model (ISEDHMM). The SM is an plug-in replacement for n-gram
models. The ISEDHMM is a plug-in replacement for hidden Markov
models. Because n-gram models and HMMs are widely used, applications
for the SM and the ISEDHMM abound. As both the SM and ISEDHMM are
Bayesian nonparametric models, a brief introduction to Bayesian
nonparametric modeling will be given. Experience gained from using SM
and ISEDHMM in a number of applied settings will be shared as well.
BIOGRAPHY:
Dr. Wood (http://www.stat.columbia.edu/~fwood) is an assistant
professor of statistics at Columbia University whose research lives at
the intersection of computer science, statistics, neuroscience, and
cognitive science. His current focus is on defining and working with
expressive Bayesian models that shed light on the path towards
artificial intelligence. Dr. Wood earned a B.S. degree in Computer
Science from Cornell University, has founded and run multiple
companies, and has a Ph.D. in Computer Science from Brown University.