Researchers have introduced a new framework called Contraction Homology (CH) for learning on graphs, which addresses limitations found in existing persistent homology (PH) methods. CH utilizes contraction operations, differing in expressivity from forward PH. The study also presents Hourglass Persistence, a novel descriptor that combines inclusions and contractions to enhance learnability and stability. These methods have been implemented with efficient, differentiable algorithms that show empirical improvements on graph datasets. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
RANK_REASON The item is an arXiv preprint detailing a new theoretical framework and algorithms for graph representation learning.
