Free-form Flows: Make Any Architecture a Normalizing Flow
This addresses the problem of architectural flexibility in normalizing flows for machine learning researchers, representing a novel method rather than an incremental improvement.
The authors tackled the constraint of analytical invertibility in normalizing flows by developing a training procedure with an efficient gradient estimator, enabling any dimension-preserving neural network to serve as a generative model through maximum likelihood training. They achieved excellent results in molecule generation benchmarks using E(n)-equivariant networks and competitive performance in an inverse problem benchmark with off-the-shelf ResNet architectures.
Normalizing Flows are generative models that directly maximize the likelihood. Previously, the design of normalizing flows was largely constrained by the need for analytical invertibility. We overcome this constraint by a training procedure that uses an efficient estimator for the gradient of the change of variables formula. This enables any dimension-preserving neural network to serve as a generative model through maximum likelihood training. Our approach allows placing the emphasis on tailoring inductive biases precisely to the task at hand. Specifically, we achieve excellent results in molecule generation benchmarks utilizing $E(n)$-equivariant networks. Moreover, our method is competitive in an inverse problem benchmark, while employing off-the-shelf ResNet architectures.