Finite-width executions of Netsor tensor programs converge to their infinite-width Gaussian-process limits in Wasserstein distance at rate O(1/√n) per hidden width, covering weight-sharing architectures including RNNs and attention.
Optimal transport -- Old and new , volume =
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2representative citing papers
Under a Bayesian posterior and local Kähler assumptions, the mixed Wirtinger derivative of the complex normalizing flow's log-determinant equals the expected difference of Ricci curvatures, yielding an averaged Kähler-Einstein-type flow.
citing papers explorer
-
Quantitative Gaussian-Process limits of Tensor Programs
Finite-width executions of Netsor tensor programs converge to their infinite-width Gaussian-process limits in Wasserstein distance at rate O(1/√n) per hidden width, covering weight-sharing architectures including RNNs and attention.
-
Complex normalizing flows can almost be information K\"ahler-Ricci flows
Under a Bayesian posterior and local Kähler assumptions, the mixed Wirtinger derivative of the complex normalizing flow's log-determinant equals the expected difference of Ricci curvatures, yielding an averaged Kähler-Einstein-type flow.