Discovering Data Structures - Nearest Neighbor Search and Beyond

Explore a seminar presentation from the Harvard CMSA New Technologies in Mathematics series where Mila researcher Omar Salemohamed delves into the fascinating intersection of deep learning and data structure discovery. Learn how neural networks can autonomously develop sophisticated data structures, with a focus on nearest neighbor search applications. Understand a novel framework that adapts to data distributions while offering precise control over query and space complexity. Discover how the framework successfully recreates classical algorithms like binary search and develops structures similar to k-d trees and locality-sensitive hashing. Examine the application of this approach to high-dimensional data representation and frequency estimation in data streams. Gain insights into the potential future applications and challenges of end-to-end learned data structures through this comprehensive 57-minute exploration of cutting-edge algorithmic research.