Timezone: »

 
Spotlight
Maximum Likelihood Training of Score-Based Diffusion Models
Yang Song · Conor Durkan · Iain Murray · Stefano Ermon

@
Score-based diffusion models synthesize samples by reversing a stochastic process that diffuses data to noise, and are trained by minimizing a weighted combination of score matching losses. The log-likelihood of score-based diffusion models can be tractably computed through a connection to continuous normalizing flows, but log-likelihood is not directly optimized by the weighted combination of score matching losses. We show that for a specific weighting scheme, the objective upper bounds the negative log-likelihood, thus enabling approximate maximum likelihood training of score-based diffusion models. We empirically observe that maximum likelihood training consistently improves the likelihood of score-based diffusion models across multiple datasets, stochastic processes, and model architectures. Our best models achieve negative log-likelihoods of 2.83 and 3.76 bits/dim on CIFAR-10 and ImageNet $32\times 32$ without any data augmentation, on a par with state-of-the-art autoregressive models on these tasks.

Author Information

Yang Song (Stanford University)
Conor Durkan (University of Edinburgh)
Iain Murray (University of Edinburgh)

Iain Murray is a SICSA Lecturer in Machine Learning at the University of Edinburgh. Iain was introduced to machine learning by David MacKay and Zoubin Ghahramani, both previous NIPS tutorial speakers. He obtained his PhD in 2007 from the Gatsby Computational Neuroscience Unit at UCL. His thesis on Monte Carlo methods received an honourable mention for the ISBA Savage Award. He was a commonwealth fellow in Machine Learning at the University of Toronto, before moving to Edinburgh in 2010. Iain's research interests include building flexible probabilistic models of data, and probabilistic inference from indirect and uncertain observations. Iain is passionate about teaching. He has lectured at several Summer schools, is listed in the top 15 authors on videolectures.net, and was awarded the EUSA Van Heyningen Award for Teaching in Science and Engineering in 2015.

Stefano Ermon (Stanford)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors