Scalable and Sample-Efficient Active Learning for Graph-Based Classification

Kevin Miller (University of California, Los Angeles)

Active learning in semi-supervised classification involves introducing additional labels for unlabelled data to improve the accuracy of the underlying classifier. A challenge is to identify which points to label to best improve performance while limiting the number of new labels; this is often reflected in a tradeoff between exploration and exploitation, similar to the reinforcement learning paradigm. I will talk about my recent work designing scalable and sample-efficient active learning methods for graph-based semi-supervised classifiers that naturally balance this exploration versus exploitation tradeoff. While most work in this field today focuses on active learning for fine-tuning neural networks, I will focus on the low-label rate case where deep learning methods are generally insufficient for producing meaningful classifiers.  

Kevin Miller is a rising 5th year Ph.D. candidate in Applied Mathematics at the University of California, Los Angeles (UCLA), studying graph-based machine learning methods with Dr. Andrea Bertozzi. He is currently supported by the DOD’s National Defense Science and Engineering Graduate (NDSEG) Fellowship and was previously supported by the National Science Foundation's NRT MENTOR Fellowship. His undergraduate degree was in Applied and Computational Mathematics from Brigham Young University, Provo. His research focuses on active learning and uncertainty quantification in graph-based semi-supervised classification.