Rob Brekelmans

University of Southern California
Information Sciences Institute,
4670 Admiralty Way
Marina del Rey, CA 90292
brekelma at usc


Fourth year PhD student at USC working with Greg Ver Steeg and Aram Galstyan using information theory for unsupervised machine learning. Most of my thinking revolves around rate-distortion and the Information Bottleneck method. I've worked on a noise model to exactly calculate the mutual information rate with neural network encoders, and applied these ideas in the context of invariant and disentangled representation learning.