Thrilled to share our recent work: Learning Latent Space Energy-Based Prior Model (arxiv.org/pdf/2006.08205…). Joint work with Bo Pang and Erik Nijkamp Erik Nijkamp. We build EBM on the latent space of the generator and learn it through prior and posterior Langevin Dynamics.

Glad to share our recent ECCV work on learning multi-layer latent variable model via short run MCMC. The model is also deeply rooted in our previous work on Alternating Back-Propagation (ABP) arxiv.org/abs/1606.08571. Both works studied the MLE learning of the generator model.

Excited to share our oral presented work in NeurIPS2022 "Adaptive Multi-stage Density Ratio Estimation for Learning Latent Space Energy-based Model": arxiv.org/abs/2209.08739. Latent EBM is learned through multiple stages of density ratios via NCE (no mcmc). #NeurIPS2022 #ebms

Check this out - our new ICML24 paper about adversarial energy-based model. The EBM can be better and more efficiently learned with diffusion!