Reproduce “Incorporating Inductive Biases to Energy-based Generative Models” 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
Yukun Li, Li-Ping Liu
The authors add a fixed, interpretable statistic with a learned weight to a neural energy model’s training objective, making known properties explicit during fitting. Exact matching of expected and data statistics holds at a likelihood maximum with neural parameters fixed; score-matching training aims approximately at this. On QM9, EDP-GNN validity rises from 88.33% to 94.52% over 10,000 samples, without guaranteeing universal validity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.