Reproduce “Wasserstein Parallel Transport for Predicting the Dynamics of Statistical Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tristan Luca Saidi, Gonzalo Mena, Larry Wasserman, Florian Gunsilius
This research introduces a framework for predicting the dynamics of statistical systems described by evolving probability distributions, leveraging "Wasserstein Parallel Transport" on the Wasserstein manifold. The method accurately captures changes in distributional shape, outperforming baselines on synthetic Gaussian data and successfully imputing gene expression dynamics in single-cell RNA sequencing datasets.
Key insights
- The Riemannian geometry of the Wasserstein manifold provides a principled way to define a distributional analogue of "parallel trends" for dynamics, overcoming the limitations of Euclidean vector space assumptions.
- Approximating parallel transport on the infinite-dimensional Wasserstein manifold can be achieved by locally relating it to parallel transport on the underlying base manifold, which simplifies the computational challenge.
- The "Wasserstein Parallel Trends" assumption not only extends classical causal inference principles to full probability distributions but also recovers the mean-based parallel trends assumption as a special case.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.