Hanna M. Pasula


Postdoctoral Associate for Professor Leslie Pack Kaelbling
(in the Computer Science and Artificial Intelligence Laboratory of the Massachussetts Institute of Technology)

Office: 32-G480 (The Stata Center, Building 32 - 32 Vassar Street, room G480)
Phone: (617) 253-7781
E-mail: pasula@csail.mit.edu


Research

My research focuses on probabilistic artificial intelligence; more specifically, on representations that combine first-order logic with probability theory.

In my dissertation, I tackled the problem of Identity Uncertainty, which occurs when an agent is given a set of observations---or knowledge base entries---that may or may not refer to the same object, and asked to reason about the set of objects actually responsible for them. To represent identity uncertain domains, I generalized conventional probabilistic relational representations by removing the assumptions that the constant symbols used are all unique, and extending the probability distribution to range over the possible ways in which these symbols may be grouped into clusters representing actual objects. I then developed an approximate inference algorithm based on Markov Chain Monte Carlo (MCMC). This approach appeared to work well in practice, and in some interesting cases I could even prove its tractability. My main applications were vehicle tracking and citation clustering.

More recently, I have been working on learning probabilistic relational models of world dynamics, as represented using an extension of probabilistic STRIPS rules. In collaboration with Professor Leslie Kaelbling and Luke Zettlemoyer, I have developed a learning algorithm based on a search through the space of rule sets, where new rules are proposed using heuristic operators inspired by inductive logic programming techniques.

A more detailed research statement can be found here. And here is my CV.

Major Publications

  • Hanna M. Pasula, Stuart Russell, Michael Ostland, and Ya'acov Ritov, Tracking many objects with many sensors. In Proc. IJCAI-99, Stockholm, 1999.
  • Hanna M. Pasula and Stuart Russell Approximate Inference For First-Order Probabilistic Languages. In Proc. IJCAI-01, Seattle, 2001.
  • Michael Shilman, Hanna M. Pasula, Stuart Russell, and Richard Newton, Statistical Visual Language Models for Ink Parsing. In Proc. AAAI Spring Symposium on Sketch Understanding, Stanford, March 2002.
  • Bhaskara Marthi, Hanna M. Pasula, Stuart Russell, Yuval Peres, Decayed MCMC Filtering. In Proc. UAI-02, Edmonton, Alberta: Morgan Kaufmann, 2002.
  • Hanna M. Pasula, Bhaskara Marthi, Brian Milch, Stuart Russell, and Ilya Shpitser, Identity Uncertainty and Citation Matching. Advances in Neural Information Processing Systems 15 (NIPS 2003). Cambridge, MA: MIT Press.
  • Hanna M. Pasula, Luke S. Zettlemoyer, and Leslie Pack Kaelbling, Learning Probabilistic Planning Rules. International Conference on Automated Planning and Scheduling, 2004.
  • Luke S. Zettlemoyer, Hanna M. Pasula, and Leslie Pack Kaelbling, Learning Planning Rules in Noisy Stochastic Worlds. Proceedings of the Twentieth National Conference on Artificial Intelligence (AAAI-05), 2005.

    Dissertation

  • Identity Uncertainty