Faculty • Affiliated Faculty • Research Associates • Graduate Students

 

Jefferson Provost

Office:

Taylor 4.134

Email:

jp@cs.utexas.edu

Homepage:

cs.utexas.edu/users/jp/

Faculty Advisors:

Ben Kuipers and Risto Miikkulainen

Research Interests

My broad interest is using machine learning to create grounded representations for Artificial Intelligence. In particular, I am investigating how a robot can learn representations of its world, as viewed through its sensorimoter system, which are useful for navigation and other higher level cognitive processes.

Projects

Learning from Uninterpreted Sensor and Effectors: How can an embodied agent or robot learn to interpret its senses and use its motor outputs to build a useful model of the world about which it can reason? I am working on unsupervised and semi-supervised methods for extracting hierarchies of useful features from uninterpreted robotic sensorimotor streams.

Learning View Prototypes for Robot Navigation: Using self-organizing neural networks, our system was able to learn a set of distinctive prototypes of sensor images of simualted robotic environment. The prototypes may be used as "views" in the causal level of the Spatial Semantic Hierarchy.

Learning Automatic Classification of Email: Recent growth in the use of email for communication and the corresponding growth in the volume of email received has made automatic processing of email desirable. Two learning methods, naive bayesian learning with bag-valued features and the RIPPER rule-learning algorithm have shown promise in other text categorization tasks. I present three experiments in automatic mail foldering and spam filtering, showing that naive bayes outperforms RIPPER in classification accuracy.

Intelligent Robotics Lab

Neural Networks Research Group