Position:  PhD Student

Current Institution:  Massachusetts Institute of Technology

Abstract:  Negatively Dependent Measures for Machine Learning

As machine learning becomes ubiquitous, so do the problems that plague it: inefficient use of data and computational power, time-consuming brute-force search through hyper-parameter space, lack of strong generalization guarantees, and more.  My research focuses on analyzing negatively dependent measures as a tool to approach these fundamental problems. These measures, among which Determinantal Point Processes have already proven to be of significant interest to the machine learning community, encode negative dependence between items in subsets of a ground set: they endow the space with repulsive forces between similar points, enabling a careful balancing of the quality and diversity of a subset. My research focuses on two key problems: (1) developing scalable learning and sampling for negatively dependent measures over large datasets, and (2) leveraging their properties to guide the search in machine learning design and analysis.

Bio:
Zelda is beginning her fifth year as a PhD student in the Department of Electrical Engineering and Computer Science (EECS) at MIT, where she studies the theory and application of negatively dependent measures for machine learning model design and optimization.  She received the 2018 Google PhD Fellowship in machine learning and was previously funded by the Criteo research faculty fellowship award.  She also interned at Google (Brain Research & Machine Intelligence), where she studied problems related to time series prediction and Determinantal Point Processes.  Previously, she received bachelor’s  and master’s degrees  from École Polytechnique in France, where she was accepted to the Corps des mines after graduation.  She was also a silver medalist at the SWERC programming competition.