For uniform keys on the d-dimensional sphere, softmax attention becomes selective at inverse temperature scaling β_n* ≍ n^{2/(d-1)}, with explicit limiting laws for attention weights and outputs in each regime.
Title resolution pending
28 Pith papers cite this work. Polarity classification is still indexing.
28
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
The paper constructs minimax-bias estimators and uniformly valid confidence intervals for weighted estimands by bounding differences via parameter heterogeneity and weight distance.
A semi-parametric framework using fractional imputation and EM algorithm for estimating causal direct and indirect effects with left-censored mediators due to assay limits.
Establishes non-asymptotic Gaussian approximation bounds for federated LSA with explicit communication-heterogeneity trade-offs and introduces an online multiplier bootstrap for last-iterate inference with validity guarantees.
Introduces action-dependent order-book feedback for online market making, yielding O(sqrt(T)) high-probability regret in stochastic i.i.d. and mean-reverting settings without smoothness assumptions, and O(T^{2/3}) in the adversarial case.
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.
Proposes a novel semi-supervised estimator for risk prediction under double censoring that combines limited gold-standard labels with large-scale surrogates, proves theoretical validity, and shows efficiency gains over supervised methods in simulations and a T2D EHR application.
Risk-controlled post-processing yields a threshold-structured policy that follows the baseline except where an oracle fallback sharply reduces conditional violation risk, achieving O(log n/n) expected excess risk in i.i.d. settings and exact risk control under exchangeability.
A new directed tree structure learning framework for zero-inflated compositional nodes uses KL divergence scoring and column-stochastic transition matrices for conditional expectations, with proven consistency and finite-sample guarantees.
A conditional adaptive perturbation approach enables valid in-sample inference for machine learning-identified subgroups with nonregular boundaries via triple robustness.
A generalized Tweedie identity and moment-generating-function representation enable nonparametric recovery of full posteriors for heteroscedastic normal means with unknown variances without specifying a prior.
A correctly specified exposure map implies design-side orthogonality conditions, so the exposure radius can be estimated by GMM and tested by overidentification — rejecting the 2 km radius in the GiveDirectly experiment.
The sharp MSE bound for the ℓ1-minimum-norm interpolator under isotropic Gaussian covariates is recovered via the geometry of symmetric Gaussian polytopes, without the convex Gaussian min-max theorem.
PLANE jointly estimates latent gene positions from a target network and proxy embeddings on a larger gene set, with provably optimal channel weighting and demonstrated gains in network recovery and imputation.
Balanced designs that balance covariates (especially blocking) are asymptotically variance-optimal for binary ATE under Neyman's nonparametric model, and a CMH-based variance estimator is finite-sample conservative and asymptotically tight under local alternatives.
A cycle-counting-ratio estimator for the β-model achieves minimax-optimal MSE and consistency under the weak conditions θ_max→0 and θ_t‖θ‖₁→∞, even at network densities near log n/n.
Laplace approximation framework for quantile regression with mixed-effects and Gaussian processes using Fisher information and population curvature of expected loss instead of observed Hessian.
Coupled initial noises in diffusion models, with designed dependence but unchanged marginal Gaussians, improve generated image diversity on Stable Diffusion variants while preserving quality and alignment.
A calibration procedure yields a weighted transported average treatment effect with asymptotically valid and efficient inference when experimental data grows slower than observational data, even without positivity or correct OLS specification.
A doubly robust, asymptotically normal estimator for regression with completely missing covariates across populations, combining importance weighting and moment imputation under a sub-population shift assumption.
A new adaptive variance estimator for relative sparsity coefficients is introduced that fully utilizes the prior asymptotic normality theorem and incorporates variable selection effects.
Develops asymptotic theory and bootstrap inference for the τ-quantile of cross-sectional individual coefficient distributions in panel data under stochastic and deterministic designs.
A meta-analytic framework estimates the resilience probability of a surrogate marker to the surrogate paradox in a new study by modeling deviations from functional relationships observed in completed trials.
Predicting question-level rectification difficulty from text and allocating human labels by a square-root rule recovers most of the hybrid human–LLM survey efficiency gains without pilot data.
citing papers explorer
-
Scaling Limits of Long-Context Transformers
For uniform keys on the d-dimensional sphere, softmax attention becomes selective at inverse temperature scaling β_n* ≍ n^{2/(d-1)}, with explicit limiting laws for attention weights and outputs in each regime.
-
Robust Inference for Weighted Estimands
The paper constructs minimax-bias estimators and uniformly valid confidence intervals for weighted estimands by bounding differences via parameter heterogeneity and weight distance.
-
Evaluating causal indirect effects when mediators are left-censored by assay limit of quantification
A semi-parametric framework using fractional imputation and EM algorithm for estimating causal direct and indirect effects with left-censored mediators due to assay limits.
-
Gaussian Approximation and Multiplier Bootstrap for Federated Linear Stochastic Approximation
Establishes non-asymptotic Gaussian approximation bounds for federated LSA with explicit communication-heterogeneity trade-offs and introduces an online multiplier bootstrap for last-iterate inference with validity guarantees.
-
Online Market Making and the Value of Observing the Order Book
Introduces action-dependent order-book feedback for online market making, yielding O(sqrt(T)) high-probability regret in stochastic i.i.d. and mean-reverting settings without smoothness assumptions, and O(T^{2/3}) in the adversarial case.
-
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.
-
Semi-supervised Method for Risk Prediction with Doubly Censored EHR Data
Proposes a novel semi-supervised estimator for risk prediction under double censoring that combines limited gold-standard labels with large-scale surrogates, proves theoretical validity, and shows efficiency gains over supervised methods in simulations and a T2D EHR application.
-
Risk-Controlled Post-Processing of Decision Policies
Risk-controlled post-processing yields a threshold-structured policy that follows the baseline except where an oracle fallback sharply reduces conditional violation risk, achieving O(log n/n) expected excess risk in i.i.d. settings and exact risk control under exchangeability.
-
Structure Learning for Directed Trees with Zero-Inflated Compositional Nodes
A new directed tree structure learning framework for zero-inflated compositional nodes uses KL divergence scoring and column-stochastic transition matrices for conditional expectations, with proven consistency and finite-sample guarantees.
-
In-Sample Evaluation of Subgroups Identified by Generic Machine Learning
A conditional adaptive perturbation approach enables valid in-sample inference for machine learning-identified subgroups with nonregular boundaries via triple robustness.
-
Nonparametric f-Modeling for Empirical Bayes Inference with Unequal and Unknown Variances
A generalized Tweedie identity and moment-generating-function representation enable nonparametric recovery of full posteriors for heteroscedastic normal means with unknown variances without specifying a prior.
-
A Design-Based Approach to Testing and Inference in (Quasi-)Experiments with Spillovers
A correctly specified exposure map implies design-side orthogonality conditions, so the exposure radius can be estimated by GMM and tested by overidentification — rejecting the 2 km radius in the GiveDirectly experiment.
-
Minimum Norm Interpolation via The Local Theory of Banach Spaces: The Role of Gaussianity
The sharp MSE bound for the ℓ1-minimum-norm interpolator under isotropic Gaussian covariates is recovered via the geometry of symmetric Gaussian polytopes, without the convex Gaussian min-max theorem.
-
AI-Augmented Statistical Network Estimation with Proxy Gene Embeddings
PLANE jointly estimates latent gene positions from a target network and proxy embeddings on a larger gene set, with provably optimal channel weighting and demonstrated gains in network recovery and imputation.
-
Optimal Designs with Robust Inference for Binary Treatment Effects
Balanced designs that balance covariates (especially blocking) are asymptotically variance-optimal for binary ATE under Neyman's nonparametric model, and a CMH-based variance estimator is finite-sample conservative and asymptotically tight under local alternatives.
-
Subgraph counting estimation for the $\beta$-model in sparse networks
A cycle-counting-ratio estimator for the β-model achieves minimax-optimal MSE and consistency under the weak conditions θ_max→0 and θ_t‖θ‖₁→∞, even at network densities near log n/n.
-
Laplace Approximations for Mixed-Effects and Gaussian Process Quantile Regression
Laplace approximation framework for quantile regression with mixed-effects and Gaussian processes using Fisher information and population curvature of expected loss instead of observed Hessian.
-
Couple to Control: Joint Initial Noise Design in Diffusion Models
Coupled initial noises in diffusion models, with designed dependence but unchanged marginal Gaussians, improve generated image diversity on Stable Diffusion variants while preserving quality and alignment.
-
Transporting treatment effects by calibrating large-scale observational outcomes
A calibration procedure yields a weighted transported average treatment effect with asymptotically valid and efficient inference when experimental data grows slower than observational data, even without positivity or correct OLS specification.
-
Augmented transfer regression learning for completely missing covariates
A doubly robust, asymptotically normal estimator for regression with completely missing covariates across populations, combining importance weighting and moment imputation under a sub-population shift assumption.
-
An adaptive variance estimator for relative sparsity
A new adaptive variance estimator for relative sparsity coefficients is introduced that fully utilizes the prior asymptotic normality theorem and incorporates variable selection effects.
-
Estimation and Inference for the $\tau$-Quantile of Individual Heterogeneous Coefficient
Develops asymptotic theory and bootstrap inference for the τ-quantile of cross-sectional individual coefficient distributions in panel data under stochastic and deterministic designs.
-
A Functional-Class Meta-Analytic Framework for Quantifying Surrogate Resilience
A meta-analytic framework estimates the resilience probability of a surrogate marker to the surrogate paradox in a new study by modeling deviations from functional relationships observed in completed trials.
-
Rectification Difficulty and Optimal Sample Allocation in LLM-Augmented Surveys
Predicting question-level rectification difficulty from text and allocating human labels by a square-root rule recovers most of the hybrid human–LLM survey efficiency gains without pilot data.
-
Learning Mixtures of Nonparametric and Convolutional Measures on Effectively Low-dimensional Affine Spaces
Mixtures of convolutional measures on low-dimensional affine spaces admit unique identifiability in semi-parametric settings and posterior contraction rates under convex polytope support assumptions in a well-specified Bayesian regime.
-
Rectified Flow: A Marginal Preserving Approach to Optimal Transport
A single-objective rectified flow variant uses neural ODEs trained by regression to monotonically decrease a fixed convex transport cost while preserving marginal distributions.
-
Generating Plausible Stress Scenarios via Large Deviations
A large-deviations method generates plausible stress scenarios for financial losses by concentrating on most likely configurations conditional on large losses, recovering stressed loss laws even with sparse data.
-
Explainable AI Isn't Enough! Rethinking Algorithmic Contestability
The paper defines algorithmic contestability as identifying evidence to overturn potentially incorrect decisions and identifies three types of such evidence that make decisions normatively indefensible under the decision maker's standards.