THESIS DEFENSE: Probabilistic programs, random data structures and the computability of distributions
Speaker: Daniel Roy , CSAIL MITContact:
Date: March 9 2011
Time: 9:30AM to 10:30AM
Location: G449 Patil/Kiva
Host: Leslie Kaelbling, CSAIL MIT
Daniel Roy, 617 872 3267, firstname.lastname@example.org
How much of probabilistic inference can be automated?
We investigate the class of probability distributions that can be represented by *algorithms* and explore the fundamental limitations of using this representation to describe and compute conditional distributions. In addition to results showing the nonexistence of generic inference algorithms (even inefficient ones), we highlight some positive results showing that posterior inference is possible in the presence of additional structure like exchangeability and noise.
This theoretical work bears on the development of *probabilistic programming languages* (which enable the specification of complex, recursively defined, probabilistic models) and the design of Monte Carlo-based compilers (which can be used to transform algorithmic representations of a joint distribution into algorithms for sampling from its associated conditional distributions).
Thesis Advisor: Leslie P. Kaelbling
Thesis Committee : Joshua B. Tenenbaum, Scott Aaronson, Yee Whye Teh
See other events that are part of
See other events happening in March 2011