A General Agnostic Active Learning Algorithm

We present an agnostic active learning algorithm for any hypothesis class of bounded VC dimension under arbitrary data distributions. Most previous work on active learning either makes strong distributional assumptions, or else is computationally prohibitive. Our algorithm extends the simple scheme of Cohn, Atlas, and Ladner [1] to the agnostic setting, using reductions to supervised learning that harness generalization bounds in a simple but subtle manner. We provide a fall-back guarantee that bounds the algorithm's label complexity by the agnostic PAC sample complexity. Our analysis yields asymptotic label complexity improvements for certain hypothesis classes and distributions. We also demonstrate improvements experimentally.

[1]  Vladimir Vapnik,et al.  Chervonenkis: On the uniform convergence of relative frequencies of events to their probabilities , 1971 .

[2]  Gábor Lugosi,et al.  Introduction to Statistical Learning Theory , 2004, Advanced Lectures on Machine Learning.

[3]  H. Sebastian Seung,et al.  Selective Sampling Using the Query by Committee Algorithm , 1997, Machine Learning.

[4]  David A. Cohn,et al.  Improving generalization with active learning , 1994, Machine Learning.

[5]  Sanjoy Dasgupta,et al.  Coarse sample complexity bounds for active learning , 2005, NIPS.

[6]  Prasad Raghavendra,et al.  Hardness of Learning Halfspaces with Noise , 2006, 2006 47th Annual IEEE Symposium on Foundations of Computer Science (FOCS'06).

[7]  Claire Monteleoni,et al.  Learning with online constraints: shifting concepts and active learning , 2006 .

[8]  Claire Monteleoni,et al.  Efficient Algorithms for General Active Learning , 2006, COLT.

[9]  Matti Kääriäinen,et al.  Active Learning in the Non-realizable Case , 2006, ALT.

[10]  Steve Hanneke,et al.  Teaching Dimension and the Complexity of Active Learning , 2007, COLT.

[11]  Maria-Florina Balcan,et al.  Margin Based Active Learning , 2007, COLT.

[12]  Steve Hanneke,et al.  A bound on the label complexity of agnostic active learning , 2007, ICML '07.

[13]  Robert D. Nowak,et al.  Minimax Bounds for Active Learning , 2007, IEEE Transactions on Information Theory.

[14]  Adam Tauman Kalai,et al.  Analysis of Perceptron-Based Active Learning , 2009, COLT.

[15]  John Langford,et al.  Agnostic active learning , 2006, J. Comput. Syst. Sci..