Reproduce “Persistence Positional Encoding: A Minimal Transformer Recipe for Persistence Diagrams” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Eunwoo Heo, Ganghyun Kim, Jaejun Yoo
The authors introduce PPE, which represents each persistence-diagram point as a token for a standard Transformer. It maps birth-and-death coordinates into fixed random Fourier features, adds a homology embedding, and uses a classification token without positional encoding. On TU8, fixed features reach 72.63% accuracy, compared with 67.14% for a learned point MLP, a controlled comparison among tested choices.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.