FFJORD: Free-form Continuous Dynamics for Scalable Reversible Generative Models
This addresses the scalability and flexibility limitations in reversible generative models for researchers and practitioners in machine learning, representing a novel method rather than an incremental improvement.
The paper tackled the problem of training invertible generative models by introducing FFJORD, which uses Hutchinson's trace estimator to enable unbiased density estimation with unrestricted neural network architectures, achieving state-of-the-art results in high-dimensional density estimation, image generation, and variational inference.
A promising class of generative models maps points from a simple distribution to a complex distribution through an invertible neural network. Likelihood-based training of these models requires restricting their architectures to allow cheap computation of Jacobian determinants. Alternatively, the Jacobian trace can be used if the transformation is specified by an ordinary differential equation. In this paper, we use Hutchinson's trace estimator to give a scalable unbiased estimate of the log-density. The result is a continuous-time invertible generative model with unbiased density estimation and one-pass sampling, while allowing unrestricted neural network architectures. We demonstrate our approach on high-dimensional density estimation, image generation, and variational inference, achieving the state-of-the-art among exact likelihood methods with efficient sampling.