Reproduce “Training Deep Energy-Based Models with f-Divergence Minimization” 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

Lantao Yu, Yang Song, Jiaming Song, Stefano Ermon

f-EBM trains energy-based models with different distribution-mismatch objectives by combining the model’s energy with a learned auxiliary function. Its variational formulation supports alternating updates on real and generated samples without differentiating through the sampling chain. On conditional CIFAR-10 generation, Jensen–Shannon f-EBM reports Inception Score 8.61 ± 0.06 and FID 30.86, versus 8.30 and 37.90 for CD (KL); gains are task-specific, not universal.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.