Riemannian Generative Decoder

TL;DR

The paper introduces a Riemannian generative decoder that learns manifold-valued latents without an encoder, simplifying training and respecting data geometry, validated on diverse case studies.

Contribution

It proposes a unified approach for learning manifold-valued latents directly with a decoder, avoiding brittle density estimation and broadening applicability to various Riemannian manifolds.

Findings

Learned representations respect prescribed geometry.

Method captures intrinsic non-Euclidean structure.

Compatible with existing architectures and yields interpretable latents.

Abstract

Euclidean representations distort data with intrinsic non-Euclidean structure. While Riemannian representation learning offers a solution by embedding data onto matching manifolds, it typically relies on an encoder to estimate densities on chosen manifolds. This involves optimizing numerically brittle objectives, potentially harming model training and quality. To completely circumvent this issue, we introduce the Riemannian generative decoder, a unifying approach for finding manifold-valued latents on any Riemannian manifold. Latents are learned with a Riemannian optimizer while jointly training a decoder network. By discarding the encoder, we vastly simplify the manifold constraint compared to current approaches which often only handle few specific manifolds. We validate our approach on three case studies -- a synthetic branching diffusion process, human migrations inferred from mitochondrial DNA, and cells undergoing a cell division cycle -- each showing that learned representations respect the prescribed geometry and capture intrinsic non-Euclidean structure. Our method requires only a decoder, is compatible with existing architectures, and yields interpretable latent spaces aligned with data geometry. Code available on https://github.com/yhsure/riemannian-generative-decoder.