Awarded in 2017
Home Department: Computer Science
Faculty Advisors: Christopher Ré (Computer Science) and Daniel Rubin (Biomedical Data Science, Radiology, Medicine - Biomedical Informatics Research)
Research Title: Alleviating the Labeling Bottleneck with Weak Supervision
Research Description: Recent advances in machine learning have led to impressive successes in many domains; however, these methods rely on massive hand-labeled training sets. Alex’s research addresses this “labeling bottleneck” by enabling users to train machine learning models with higher level, less precise inputs, and by easily leveraging structured data resources available in domains such as bioinformatics. In the first part of his PhD, Alex developed data programming, a method for using lower-accuracy rules to train high-accuracy models, and Snorkel, a system which uses this paradigm for extracting information from text documents such as scientific articles and electronic health records. In his subsequent work, Alex seeks to develop similar techniques that extend to image and video data, focused initially on improving the performance of automated disease diagnosis using radiology and histopathology data.