![]() |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Machine Learning Seminar Series | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Date: | September 15, 2008 |
| Time: | 4:30 PM - 00:00 AM () |
| Location: | Lower Level, CIC Bldg. Other |
| Speaker: | John Langford Director of Learning, Yahoo! Research |
| Title: | Importance Weighted Active Learning |
| Abstract: | I will present a family of active learning algorithms which is sound, practical, and broadly applicable. In more detail, the algorithm: (a) has provably bounded label complexity, even in the presence of adversarial noise. (b) yields computationally tractable algorithms, which work in practice. (c) applies to essentially all loss functions. (Joint work with Sanjoy Dasgupta and Alina Beygelzimer) |
