Learning High Dimensional Wasserstein Geodesics
arXiv:2102.02992
Abstract
We propose a new formulation and learning strategy for computing the Wasserstein geodesic between two probability distributions in high dimensions. By applying the method of Lagrange multipliers to the dynamic formulation of the optimal transport (OT) problem, we derive a minimax problem whose saddle point is the Wasserstein geodesic. We then parametrize the functions by deep neural networks and design a sample based bidirectional learning algorithm for training. The trained networks enable sampling from the Wasserstein geodesic. As by-products, the algorithm also computes the Wasserstein distance and OT map between the marginal distributions. We demonstrate the performance of our algorithms through a series of experiments with both synthetic and realistic data.
References in corpus (11)
- Adam: A Method for Stochastic Optimization
- Spectral Normalization for Generative Adversarial Networks
- Near-linear time approximation algorithms for optimal transport via Sinkhorn iteration
- A Machine Learning Framework for Solving High-Dimensional Mean Field Game and Mean Field Control Problems
- Wasserstein Distributionally Robust Stochastic Control: A Data-Driven Approach
- How to train your neural ODE: the world of Jacobian and kinetic regularization
- A Fast Proximal Point Method for Computing Exact Wasserstein Distance
- Unnormalized Optimal Transport
- Scalable Computations of Wasserstein Barycenter via Input Convex Neural Networks
- Learning to Match via Inverse Optimal Transport
- Learning Stochastic Behaviour from Aggregate Data