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

  • Anqi Liu | University of Illinois at Chicago
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.
Speaker Details

Anqi Liu is a Ph.D. Candidate in the Department of Computer Science at the University of Illinois at Chicago. Her research focus is in robust machine learning under sample selection bias (covariate shift) and active learning. She is interested in making learning from various data sources more robustly and safe, as well as enabling better interaction between humans and AI systems. She received B.E. in Software Engineering and B.A. in Finance from Tianjin University of Finance and Economics in China (2012). She has been a data scientist intern at Microsoft, and a research intern at NEC Labs. Her thesis proposal received an AAAI Doctoral Consortium Scholarship.

Watch Next