Answers are grounded in this paper and cite their sources.
Minimizing functionals in the space of probability distributions can be done Wasserstein gradient flows. To solve them numerically, a possible approach to rely on the Jordan-Kinderlehrer-Otto (JKO) scheme which is analogous to proximal scheme in Euclidean spaces. However, it requires solving a nested problem at each iteration, and is known for its computational, especially in high dimension. To alleviate it, very recent works to approximate the JKO scheme leveraging Brenier's theorem, and using of Input Convex Neural Networks to parameterize the density(JKO-ICNN). However, this method comes with a high computational cost and issues. Instead, this work proposes to use gradient flows in the of probability measures endowed with the sliced-Wasserstein (SW). We argue that this method is more flexible than JKO-ICNN, since SW a closed-form differentiable approximation. Thus, the density at each can be parameterized by any generative model which alleviates the burden and makes it tractable in higher dimensions.
No takes yet. Share an insight, caveat, or question.
Bonet et al. (2021) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: