| Amin H. Atrash
School of Computer Science
Email:aatras at cs mcgill ca
Phone:514-398-7071 ext. 00141#
Office:McConnell Room 111S
|About:||I am currently a PhD student at the School of Computer Science
at McGill University. I work under Professor Joelle Pineau as part of the
Reasoning and Learning Lab and the Centre for Intelligent Machines.
I received my undergraduate degree in computer science in 1999 and my masters degree in 2002 from the Georgia Institute of Technology. From 2003 to 2005, I worked with the speech recognition group at BBN Technologies.
|Research Summary:||My research interests focus on reasoning and learning algorithms and
their application to real world domains. In particular, my work focuses on
Markov models and their use on robot platforms.
|Research Projects:||SmartWheeler -
The SmartWheeler project takes advantage of advancements in robot
technology to develop an autonomous wheelchair for use by disabled patients who
might normally have difficultly operating a standard electric wheelchair. Work
on this project focuses on development of an intelligent interaction manager to
allow the user to control the wheelchair through a multimodal interface,
particularly speech. The dialogue management is handled by a partially
observable Markov decision process (POMDP) which determines the correct action,
such as issuing a command to the robot, requesting clarification, or returning
information to the user. This dialogue management system allows for a more
natural interaction with the wheelchair beyond simple voice commands. This
project includes construction and maintenance of the robot platform,
development of the dialogue management system, and user studies on patients in
a rehabilitation center.
POMDP Learning - POMDP research has typically been focused on planning using existing POMDP models. The goal of this project is to use active learning to discover the parameters of the model itself. By requesting information from an oracle which has the optimal desired policy, a POMDP model is refined over time to better approximate the real world. Using data from this oracle, the model can adapt parameters which cause the resulting policy to behave as the oracle.
|Teaching:||COMP 526 - Probabilistic AI