Reproduce “Learning the Geometry of Materials Space with Trained Neural Tangent Kernels” 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
Daniel Kaplan, Gabriel Kotliar, Anirvan M Sengupta
The authors examine the trained neural tangent kernel of a composition-to-superconducting-temperature predictor as a materials similarity map. Its neighborhoods reflect parameter-response similarity, not necessarily shared elements or similar transition temperatures. Trained-kernel regression approaches network errors (2.24 K and 7.61 K versus 2.11 K and 7.20 K), unlike untrained-kernel regression (21.7 K and 31.9 K), with no independent test split specified.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.