Research Project Title:
Learned Bloom Filters
abstract:MIT's Data Science and AI Lab (DSAIL) has shown that optimizing data structures with machine learning can have significant performance benefits over their generic counterparts in databases. For Bloom Filter structures this meant reducing space use by anywhere from 15-40%. However, because the benefits vary by use case and evaluating them requires mastery in NLP and Machine Learning, those benefits remain unharnessed. The goal of my SuperUROP is to build a general purpose learned structures toolkit such that others can easily utilize and evaluate the Learned Bloom Filter.
“SuperUROP is an opportunity for me to become an expert in a developing area of the ML field. I hope to publish a paper, and use this work to transition to graduate level. My previous experience includes advanced ML coursework (6.867 and 6.819), teaching as a lab assistant for Intro to Machine Learning (6.036), and professional software development. I have high hopes for this project’s impact on programming efficiency.”