Machine Learning on Point Clouds and Other Varying-Size Objects

Learn about dimension-independent machine learning architectures for handling inputs of varying sizes in this 52-minute conference talk. Explore a novel invariant machine learning model for point clouds developed using concepts from Galois theory, and discover how this approach addresses the challenge of processing data structures like graphs, sets, and point clouds that don't have fixed dimensions. Examine a comprehensive framework for transferability across different dimensions, where you'll understand how transferability relates to continuity in limit spaces created by identifying small problem instances with their equivalent large counterparts. Gain insights into the theoretical foundations that enable machine learning models to work effectively across varying input sizes, bridging concepts from theoretical computer science and practical machine learning applications.