A Learning-based Algorithm Selection Meta-reasoner for the Real-time MPE Problem

Speaker:        Dr. Haipeng Guo
                Department of Computer Science
                The Hong Kong University of Science & Technology

Title:          "A Learning-based Algorithm Selection Meta-reasoner
                for the Real-time MPE Problem"

Date:           Wednesday, 30 June 2004

Time:           4:15 pm - 5:15 pm

Venue:          Lecture Theatre F
                (Leung Yat Sing Lecture Theatre, near lift nos. 25/26)
                HKUST


ABSTRACT:

The algorithm selection problem aims to select the best algorithm for an 
input problem instance according to some characteristics of the instance. 
The general algorithm selection problem is undecidable therefore 
analytical methods are not applicable. However, it is possible to learn a 
predictive algorithm selection model from empirical algorithm performance 
data. This talk presents a machine learning-based inductive approach to 
build such an algorithm selection meta-reasoner for the real-time Most 
Probable Explanation (MPE) problem. The experimental results show that 
the learned algorithm selection models can help integrate multiple MPE 
algorithms to gain a better overall performance of reasoning.

 
*******************

Biography:

Dr. Haipeng Guo received his Ph.D. in Department of Computing & 
Information Science, Kansas State University, in 2003. Research interests 
include Bayesian networks, probabilistic inference, machine learning, 
NP-hard optimization, etc.