Implicit generation and generalization methods for energy-based models
OpenAI Blog
Researchers improved the training stability and scalability of energy-based models to achieve better sample quality and generalization than existing approaches. The method uses iterative refinement during generation and produces samples competitive with GANs while maintaining the mode coverage guarantees of likelihood-based models. This work may encourage further development of energy-based models as an alternative generative modeling approach.
Why it matters
We’ve made progress towards stable and scalable training of energy-based models (EBMs) resulting in better sample quality and generalization ability than existing models. Generation in EBMs spends more compute to continually refine its answers and doing so can generate samples competitive with GANs at low temperatures, while also having mode coverage guarantees of likelihood-based models. We hope these findings stimulate further research into this promising class of models.
