Experience with Markov Logic Networks in a Large AI System
Speaker: Thomas G. Dietterich , Oregon State University
Date: July 16 2007
Time: 2:00PM to 3:00PM
Location: Kiva, 32-G449
Host: Professor Leslie Kaelbling, MIT CSAIL
Contact: Teresa Cataldo, 617-452-5005, cataldo@csail.mit.edu
Relevant URL: CALO is an integrated AI system that seeks to provide support for the
modern knowledge worker. Led by SRI, the CALO project includes
contributions from more than 25 research groups in the U.S. To
integrate the various learning components, and to combine hand-written
probabilistic rules with factual and learned knowledge, we implemented
and deployed a Markov Logic system that we call the Probabilistic
Constraint Engine (PCE). This talk will describe our experiences
with the PCE in CALO. The PCE provides several functions. First, it
integrates probabilistic predictions from various learned and
hand-authored components to maintain a relational model of the user's
work environment (projects, appointments, action items, files,
folders, email messages, email folders, web pages, etc.). Second, it
provides a general mechanism for probabilistic credit assignment, so
that corrective feedback from the user can be routed back to the
classifier responsible for an incorrect prediction. Third, it
provides a mechanism for various forms of co-training and
semi-supervised learning. The talk will also discuss future
directions for the PCE.
See other events that are part of Language, Learning, Vision and Graphics Seminar Series (LLVG) 2006/2007
See other events happening in July 2007
