Avoiding the Pitfalls of Active Learning with Robust Predictors for Covariate Shift

Published on 13 Mar 2018, 19:05
Pool-based active learning promises to significantly reduce the labeling burden of black-box supervised machine learning methods but often doesn't deliver in practice. In fact, standard active learning techniques frequently provide worse performance than passive learners (i.e., datapoints labeled at random). This talk will illuminate the fundamental issue of covariate shift hindering pool-based active learning methods, present a new approach using adversarial estimation for addressing it, demonstrate the benefits of the approach on classification tasks, and discuss extensions of this idea for other prediction tasks.

See more at microsoft.com/en-us/research/video/avoid...
news tech music