Prototype Selection Using Topological Data Analysis
Researchers have introduced two novel prototype selection methods, Topological Prototype Selector (TPS) and Boundary-Conscious Topological Prototype Selector (BoundaryTPS), which leverage topological data analysis. These methods operate on the multi-scale topological structure of data, unlike traditional approaches. Evaluations on fifteen real datasets showed that these topological methods offer a different operating point in prototype selection, with BoundaryTPS outperforming several classical baselines in preserving H1 persistence diagrams and demonstrating greater stability under perturbation. AI
IMPACT Introduces novel techniques for data compression and representation that could improve efficiency in machine learning pipelines.