The test error of random-feature ridge regression with arbitrary data augmentation admits a closed-form asymptotic characterization in the proportional regime that depends only on population covariances and augmentation statistics.
Title resolution pending
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
years
2026 8representative citing papers
The Sinkhorn treatment effect is a new entropic optimal transport measure of divergence between counterfactual distributions that admits first- and second-order pathwise differentiability, debiased estimators, and asymptotically valid tests for distributional treatment effects.
PUICL is a transformer pretrained on synthetic PU data from structural causal models that solves positive-unlabeled classification via in-context learning without gradient updates or fitting.
Perturb-and-Correct generates epistemically diverse predictors from a single pretrained network via hidden-layer perturbations followed by affine least-squares corrections that enforce agreement on calibration data.
Resolvents of the sample covariances in the separable mixture model approximate deterministic matrices defined via solutions to a dual system of equations, without simultaneous diagonalizability assumptions.
Dynamic directed spectral co-clustering on degree-corrected stochastic co-blockmodels embedded in VAR-type models uncovers latent community paths, with non-asymptotic misclassification bounds and applications to U.S. payrolls and global stock volatilities.
A generalized zeroth-order method samples random directions on the sphere to optimize quotients of quadratics, estimates Riemannian derivatives with surrogates, and yields an accelerated algorithm outperforming prior work.
In finite-depth random linear optical circuits, entanglement grows at most diffusively and robust circuit complexity scales similarly, with depth bounds ensuring near-maximal subsystem entanglement and closeness to Haar unitaries.
citing papers explorer
-
Characterizing the Generalization Error of Random Feature Regression with Arbitrary Data-Augmentation
The test error of random-feature ridge regression with arbitrary data augmentation admits a closed-form asymptotic characterization in the proportional regime that depends only on population covariances and augmentation statistics.
-
Sinkhorn Treatment Effects: A Causal Optimal Transport Measure
The Sinkhorn treatment effect is a new entropic optimal transport measure of divergence between counterfactual distributions that admits first- and second-order pathwise differentiability, debiased estimators, and asymptotically valid tests for distributional treatment effects.
-
In-Context Positive-Unlabeled Learning
PUICL is a transformer pretrained on synthetic PU data from structural causal models that solves positive-unlabeled classification via in-context learning without gradient updates or fitting.
-
Perturb and Correct: Post-Hoc Ensembles using Affine Redundancy
Perturb-and-Correct generates epistemically diverse predictors from a single pretrained network via hidden-layer perturbations followed by affine least-squares corrections that enforce agreement on calibration data.
-
Spectral approximation for the separable covariance mixture model
Resolvents of the sample covariances in the separable mixture model approximate deterministic matrices defined via solutions to a dual system of equations, without simultaneous diagonalizability assumptions.
-
Latent community paths in VAR-type models via dynamic directed spectral co-clustering
Dynamic directed spectral co-clustering on degree-corrected stochastic co-blockmodels embedded in VAR-type models uncovers latent community paths, with non-asymptotic misclassification bounds and applications to U.S. payrolls and global stock volatilities.
-
Generalization of Zeroth-Order Method for Quotients of Quadratic Functions
A generalized zeroth-order method samples random directions on the sphere to optimize quotients of quadratics, estimates Riemannian derivatives with surrogates, and yields an accelerated algorithm outperforming prior work.
-
Entanglement and circuit complexity in finite-depth random linear optical networks
In finite-depth random linear optical circuits, entanglement grows at most diffusively and robust circuit complexity scales similarly, with depth bounds ensuring near-maximal subsystem entanglement and closeness to Haar unitaries.