Friday, March 1, 2002 - 6:30 p.m.
Title: Reinforcement Learning
Speaker: Jayprasad J. Hegde
Venue: Lecture Theatre, NCST, Juhu
Abstract: Reinforcement Learning is all about simulating the manner in
which humans would go about learning in real life. So, if you were
presented with a game of chess and you did not know how to play it, how
would you go about learning it? How would you go about learning it if you
were told that out of the six games you played only 2 resulted in a
victory? How do you figure out which game you have won? How do you figure
out which strategies were successful or which strategies are a strict
no-no?
These are some of the issues you would face when posed a problem of
learning a new concept, or in this case learning a new game. These are
exactly the same issues which a system would have to deal with if it were
to ever try and learn something using trial-and-error.
Now, consider the implications of using such a learning mechanism:
exploring the surface of Mars by learning how to navigate on-the-fly,
building a world-class checkers-playing program using GAs and NNs, learning
to acquire "soft" versions of a dictionary given a parallel corpus,
learning new strategies for computer-controlled opponents in the latest 3D
first-person shooting game... the list goes on.
If you've missed out on the Machine Learning seminar given previously, then
this would be a good time to attend the second of the Machine Learning
seminars viz. Reinforcement Learning, since, we will recap some of the
pertinent basics of Machine Learning, cover the basics of Reinforcement
Learning, discuss some of the technologies involved and then round-off with
a motivating case study.
About the Speaker: Jayprasad J. Hegde is a fitness-freak going under the
guise of a Computer Scientist in the KBCS division at NCST. He claims to be
working in the field of Natural Language Technology and have interests in
Machine Learning, but one is rarely fortunate enough to see him around and
confirm this. His only proof of existence seems to come from some of the
articles that he keeps writing rather frequently, his knack for being
available for "important" occasions like someone's birthday-treat and his
picture on the his home-page. If you have seen him in your locality please
dial KBCS and inform them. Please do not expect any rewards for doing so.