Geodesics in fibered latent spaces: A geometric approach to learning correspondences between conditions
arXiv:2005.07852
Abstract
This work introduces a geometric framework and a novel network architecture for creating correspondences between samples of different conditions. Under this formalism, the latent space is a fiber bundle stratified into a base space encoding conditions, and a fiber space encoding the variations within conditions. Furthermore, this latent space is endowed with a natural pull-back metric. The correspondences between conditions are obtained by minimizing an energy functional, resulting in diffeomorphism flows between fibers. We illustrate this approach using MNIST and Olivetti and benchmark its performances on the task of batch correction, which is the problem of integrating multiple biological datasets together.
arXiv comment36 pages, many figures. v1: Preliminary version. v2: Minor ref fix. v3: Submitted version with enhanced presentation
References in corpus (12)
- Adam: A Method for Stochastic Optimization
- Conditional Generative Adversarial Nets
- UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
- Disentangling by Factorising
- Semi-Supervised Learning with Generative Adversarial Networks
- MMD GAN: Towards Deeper Understanding of Moment Matching Network
- Hyperspherical Variational Auto-Encoders
- Domain-Adversarial Training of Neural Networks
- Variational Inference of Disentangled Latent Concepts from Unlabeled Observations
- Spherical Latent Spaces for Stable Variational Autoencoders
- Conditional out-of-sample generation for unpaired data using trVAE
- Holographic Neural Architectures