A Law of Robustness beyond Isoperimetry

Authors: Yihan Wu, Heng Huang, Hongyang Zhang

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We prove a Lipschitzness lower bound Ω( p n/p) of the interpolating neural network with p parameters on arbitrary data distributions. With this result, we validate the law of robustness conjecture in prior work by Bubeck, Li, and Nagaraj on two-layer neural networks with polynomial weights. We then extend our result to arbitrary interpolating approximators and prove a Lipschitzness lower bound Ω(n1/d) for robust interpolation. Our results demonstrate a two-fold law of robustness: i) we show the potential benefit of overparametrization for smooth data interpolation when n = poly(d), and ii) we disprove the potential existence of an O(1)-Lipschitz robust interpolating function when n = exp(ω(d)).
Researcher Affiliation Academia 1Department of Computer Science, University of Maryland at College Park 2School of Computer Science, University of Waterloo. Correspondence to: Yihan Wu <ywu42@umd.edu>, Heng Huang <heng@umd.edu>, Hongyang Zhang <hongyang.zhang@uwaterloo.ca>.
Pseudocode No The paper contains mathematical proofs and theoretical derivations but no explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any statements or links indicating that source code for the described methodology is publicly available.
Open Datasets No The paper discusses theoretical aspects of robust interpolation with 'noisy training data' and mentions 'CIFAR10' and 'Image Net' as examples of real-world datasets in a discussion of empirical observations by others, but does not specify any particular public or open dataset used for its own research.
Dataset Splits No The paper is theoretical and does not describe empirical experiments, therefore no dataset split information for validation is provided.
Hardware Specification No The paper is theoretical and does not describe empirical experiments, therefore no hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and does not describe empirical experiments, therefore no specific software dependencies with version numbers are mentioned.
Experiment Setup No The paper is theoretical and does not describe empirical experiments, therefore no specific experimental setup details like hyperparameters or training configurations are provided.