 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
|||||||||||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
|||
 |
 |
 |
 |
|
||||||||||
 |
|
|||||||||||||
 |
|
|||||||||||||
 |
 |
|
||||||||||||
| Seminar Schedule (Seminar Organizer: Prof. Ziv Bar-Joseph) | |
Machine Learning Lunch Seminar
|
Spring 2008 Seminar
Date: March 17, 2008 Time: 5:00 PM - 6:00 PM Location: 3305 Newell-Simon Hall Speaker: Stephen Hanneke Ph.D. Candidate Title: Theoretical Foundations of Active Learning Abstract: In active learning, a learning algorithm is given access to a large pool of unlabeled examples, and is allowed to request the label of any particular examples from that pool, interactively. The objective is to learn a function that accurately predicts the labels of new examples, while requesting as few labels as possible. Active learning can often significantly decrease the work load of human annotators, compared to passive learning where the examples to be labeled are chosen at random. This is of particular interest for learning tasks where unlabeled examples are available in abundance, but label information requires significant effort to obtain. In the passive learning literature, there are well-known bounds on the rates of convergence of the loss of certain estimators, as a function of the number of labeled examples observed. However, significantly less is presently known about the analogous rates in active learning: namely, the rate of convergence of the loss, as a function of the number of label requests made by an active learning algorithm. In this thesis proposal, I will outline some recent progress toward understanding convergence rate improvements achievable by active learning, along with general algorithms that achieve them. I will also describe a few of the many open problems remaining on this topic, which I hope to tackle in the near future. Committee: Eric Xing (Chair), Avrim Blum, Sanjoy Dasgupta (UCSD), Larry Wasserman. Proposal Document: http://www.cs.cmu.edu/~shanneke/docs/hanneke-proposal.pdf
