Pulling back information geometry
arXiv:2106.05367
Abstract
Latent space geometry has shown itself to provide a rich and rigorous framework for interacting with the latent variables of deep generative models. The existing theory, however, relies on the decoder being a Gaussian distribution as its simple reparametrization allows us to interpret the generating process as a random projection of a deterministic manifold. Consequently, this approach breaks down when applied to decoders that are not as easily reparametrized. We here propose to use the Fisher-Rao metric associated with the space of decoder distributions as a reference metric, which we pull back to the latent space. We show that we can achieve meaningful latent geometries for a wide range of decoder distributions for which the previous theory was not applicable, opening the door to `black box' latent geometries.
arXiv commentPresented at AISTATS 2022
References in corpus (11)
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Auto-Encoding Variational Bayes
- Optimizing Neural Networks with Kronecker-factored Approximate Curvature
- New insights and perspectives on the natural gradient method
- An elementary introduction to information geometry
- Metrics for Deep Generative Models
- Metrics for Probabilistic Geometries
- Fast and Robust Shortest Paths on Manifolds Learned from Data
- Geodesic Clustering in Deep Generative Models
- Expected path length on random manifolds
- The Fisher-Rao geometry of beta distributions applied to the study of canonical moments