Lifting Architectural Constraints of Injective Flows
This addresses computational and architectural bottlenecks for researchers and practitioners using Injective Flows to model data on lower-dimensional manifolds, representing an incremental improvement.
The paper tackles the problem of Injective Flows being limited by restrictive architectures and high computational costs by introducing a new efficient estimator for the maximum likelihood loss that is compatible with free-form bottleneck architectures, and demonstrates competitive performance on toy, tabular, and image data.
Normalizing Flows explicitly maximize a full-dimensional likelihood on the training data. However, real data is typically only supported on a lower-dimensional manifold leading the model to expend significant compute on modeling noise. Injective Flows fix this by jointly learning a manifold and the distribution on it. So far, they have been limited by restrictive architectures and/or high computational cost. We lift both constraints by a new efficient estimator for the maximum likelihood loss, compatible with free-form bottleneck architectures. We further show that naively learning both the data manifold and the distribution on it can lead to divergent solutions, and use this insight to motivate a stable maximum likelihood training objective. We perform extensive experiments on toy, tabular and image data, demonstrating the competitive performance of the resulting model.