Importance weighted active learning

We propose an importance weighting framework for actively labeling samples. This technique yields practical yet sound active learning algorithms for general loss functions. Experiments on passively labeled data show that this approach effectively reduces the label complexity required to achieve good prediction performance on many learning problems.

• Type: Conference
• Year: 2009
• Where: ICML
• Authors: Alina Beygelzimer, Sanjoy Dasgupta, John Langford
• Keywords: Active Learning Algorithms, ICML 2009, Importance Weighting Framework, Machine Learning