PILLAR Project in UTCS

Project for Interactive Learning from Language Advice and Reinforcements

The goal of the PILLAR project is to broaden the communication channel between machine learners and their human teachers. This can be achieved (1) by allowing human users to give natural language advice to help a reinforcement learning agent improve performance; and (2) by allowing agents to actively solicit advice and other forms of tutorial feedback when it is needed.

This is joint work with Prof. Jude Shavlik's research group in Department of Computer Sciences at the University of Wisconsin at Madison.

The project is supported by a grant HR0011-04-1-0007 from the DARPA Information Processing Technology Office. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of DARPA or the US Government.


The PILLAR researchers at UT meet biweekly to discuss papers in the areas of reinforcement learning and natural language learning. Our next scheduled meeting will be on 6/1/2005 (Wed) at 11:00 am in ACES 3.408, where we will discuss:

Richard Maclin, Jude Shavlik, Lisa Torrey, Trevor Walker and Edward Wild
Giving Advice about Preferred Actions to Reinforcement Learners via Knowledge-Based Kernel Regression
To appear in Proceedings of the Twentieth National Conference on Artificial Intelligence (AAAI-2005), 2005.

Previously discussed papers


Researchers at UT

Researchers at U-Wisc

Yuk Wah Wong
Last modified: Wed Feb 2 12:56:26 CST 2005