 |
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
|||||||||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
|
 |
 |
|
||||||||||
 |
|
|||||||||||
 |
|
|||||||||||
 |
 |
|
||||||||||
| Seminar Schedule (Seminar Organizer: Prof. Ziv Bar-Joseph) | |
Machine Learning Lunchtime Chats
|
Spring 2007 Seminar
Date: April 2, 2007 Time: 12:00 PM - 1:00 PM Location: 3305 Newell-Simon Hall Speaker: Charles Kemp Ph.D. Candidate Title: Probabilistic Models of Human Learning Abstract: Theories of human and machine learning have influenced each other for more than fifty years, and the interaction between these fields continues to be productive. I will discuss two probabilistic models that attempt to contribute to both fields. Standard methods for unsupervised learning discover representations of a single kind: for instance, algorithms for hierarchical clustering can only discover tree structures, and algorithms for dimensionality-reduction can only discover low-dimensional spaces. I will present a hierarchical Bayesian framework that discovers for itself which kind of representation is best for a given problem. Simple representations like trees and low-dimensional spaces are useful in some contexts, but many aspects of human knowledge demand richer relational representations. I will present a nonparametric Bayesian method for discovering simple relational theories and will show how it can be tested as a psychological model.