First population risk bounds for KANs under mini-batch DP-SGD with correlated noise, using a new non-convex optimization analysis combined with stability-based generalization.
Cambridge university press
9 Pith papers cite this work. Polarity classification is still indexing.
9
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
Random sampling maps (with-replacement, binning, species) induce metrics that give uniform any-dimensional generalization and sketching rates for continuous functions on sequences, graphs and tensors.
DAPRO adaptively spends a fixed interaction budget across multi-turn LLM conversations and builds valid lower bounds on time-to-jailbreak, with coverage error scaling as the square root of the mean censoring weight.
Establishes O(d² δ^{-3} ε^{-3}) SZO complexity to reach (δ,ε)-Goldstein stationary points in non-smooth non-convex stochastic zeroth-order optimization with decision-dependent distributions, plus improved rates for smooth and Hessian-Lipschitz cases.
For two-layer KANs trained with gradient descent under logistic loss and NTK-separable assumption, polylogarithmic width suffices for 1/T optimization and 1/n generalization rates, while differential privacy requires the same width and yields √d/(nε) utility.
A neural doubly robust proxy causal learning framework using mean embeddings for treatment bridges provides consistent estimators for causal dose-response functions under unobserved confounding for continuous and structured treatments.
Derives instance-specific lower bounds on sample complexity for rank-adaptive matrix estimation and proposes a least-squares plus universal singular-value-thresholding algorithm whose finite-sample error nearly matches those bounds.
MDMF detects AI-generated images by learning patch-level forensic signatures and quantifying their distributional discrepancies with MMD, yielding larger separation than global methods when micro-defects are present.
A robust SCQM model extending SQMF to accommodate generalized Gaussian and radial Laplace noise, solved via gradient descent with line search and validated through sensitivity analysis and experiments.
citing papers explorer
-
Population Risk Bounds for Kolmogorov-Arnold Networks Trained by DP-SGD with Correlated Noise
First population risk bounds for KANs under mini-batch DP-SGD with correlated noise, using a new non-convex optimization analysis combined with stability-based generalization.
-
Any-Dimensional Learning by Sampling
Random sampling maps (with-replacement, binning, species) induce metrics that give uniform any-dimensional generalization and sketching rates for continuous functions on sequences, graphs and tensors.
-
How Many Iterations to Jailbreak? Dynamic Budget Allocation for Multi-Turn LLM Evaluation
DAPRO adaptively spends a fixed interaction budget across multi-turn LLM conversations and builds valid lower bounds on time-to-jailbreak, with coverage error scaling as the square root of the mean censoring weight.
-
Stochastic Non-Smooth Non-Convex Optimization with Decision-Dependent Distributions
Establishes O(d² δ^{-3} ε^{-3}) SZO complexity to reach (δ,ε)-Goldstein stationary points in non-smooth non-convex stochastic zeroth-order optimization with decision-dependent distributions, plus improved rates for smooth and Hessian-Lipschitz cases.
-
Optimization, Generalization and Differential Privacy Bounds for Gradient Descent on Kolmogorov-Arnold Networks
For two-layer KANs trained with gradient descent under logistic loss and NTK-separable assumption, polylogarithmic width suffices for 1/T optimization and 1/n generalization rates, while differential privacy requires the same width and yields √d/(nε) utility.
-
Doubly Robust Proxy Causal Learning with Neural Mean Embeddings
A neural doubly robust proxy causal learning framework using mean embeddings for treatment bridges provides consistent estimators for causal dose-response functions under unobserved confounding for continuous and structured treatments.
-
Near-optimal Rank Adaptive Inference of High Dimensional Matrices
Derives instance-specific lower bounds on sample complexity for rank-adaptive matrix estimation and proposes a least-squares plus universal singular-value-thresholding algorithm whose finite-sample error nearly matches those bounds.
-
Micro-Defects Expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts
MDMF detects AI-generated images by learning patch-level forensic signatures and quantifying their distributional discrepancies with MMD, yielding larger separation than global methods when micro-defects are present.
-
Robust Subspace-Constrained Quadratic Models for Low-Dimensional Structure Learning
A robust SCQM model extending SQMF to accommodate generalized Gaussian and radial Laplace noise, solved via gradient descent with line search and validated through sensitivity analysis and experiments.