Reproduce “Transformers are efficient hierarchical chemical graph learners” 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
Zihan Pengmei, Zimu Li, Chih-chan Tien, Risi Kondor, Aaron R. Dinner
SubFormer predicts molecular properties by grouping atoms into chemically meaningful fragments. A shallow message-passing network gathers local atom information into a junction tree, and a transformer lets fragment tokens interact directly, reducing attention tokens when the tree is smaller than the molecular graph. On ZINC, it reports 0.077 ± 0.003 test MAE, versus GraphGPS’s 0.070 ± 0.004, so it is competitive but not more accurate in that comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.