A general agnostic active learning algorithm

We present a simple, agnostic active learning algorithm that works for any hypothesis class of bounded VC dimension, and any data distribution. Our algorithm extends a scheme of Cohn, Atlas, and Ladner to the agnostic setting, by (1) reformulating it using a reduction to supervised learning and (2) showing how to apply generalization bounds even for the non-i.i.d. samples that result from selective sampling. We provide a general characterization of the label complexity of our algorithm. This quantity is never more than the usual PAC sample complexity of supervised learning, and is exponentially smaller for some hypothesis classes and distributions. We also demonstrate improvements experimentally.

• Type: Conference
• Year: 2007
• Where: NIPS
• Authors: Sanjoy Dasgupta, Daniel Hsu, Claire Monteleoni
• Keywords: Agnostic Active Learning, Information Technology, NIPS 2007, PAC Sample Complexity, Supervised Learning