Efficient Testable Learning of Halfspaces with Adversarial Label Noise

Authors: Ilias Diakonikolas, Daniel Kane, Vasilis Kontonis, Sihan Liu, Nikos Zarifis

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

Reproducibility Variable Result LLM Response
Research Type Theoretical We give the first polynomial-time algorithm for the testable learning of halfspaces in the presence of adversarial label noise under the Gaussian distribution.
Researcher Affiliation Academia Ilias Diakonikolas University of Wisconsin, Madison ilias@cs.wisc.edu Daniel M. Kane University of California, San Diego dakane@ucsd.edu Vasilis Kontonis University of Texas, Austin mailto:vasilis@cs.utexas.edu Sihan Liu University of California, San Diego il046@ucsd.edu Nikos Zarifis University of Wisconsin, Madison zarifis@wisc.edu
Pseudocode Yes Input: Sample access to a distribution Dx over Rd; tolerance parameter η > 0; unit vector v Rd; failure probability τ (0, 1). Output: Certifies that for all unit vectors w such that w v 2 η it holds that Prx Dx[sign(v x) = sign(w x)] Cη, for some absolute constant C > 1, or reports that Dx is not N(0, I). 1. Set B = p log(1/η)/η . 2. Let e D be the empirical distribution obtained by drawing poly(d, 1/η) log(1/τ) samples from Dx. 3. For integers B 1 i B, define Ei to be the event that {v x [iη, (i + 1)η]} and EB+1 to be the event that {|v x| p 4. Verify that PB+1 i= B 1 Pr N(0,I) [Ei] Pr e D [Ei] η. 5. Let Si the distribution of e D conditioned on Ei and S i be Si projected on the subspace orthogonal to v. 6. For each i, verify that S i has bounded covariance, i.e., check that Ex S i [xx ] 2I . Algorithm 1: Wedge-Bound
Open Source Code No The paper does not provide any statement or link indicating that the source code for the methodology described in this paper is openly available.
Open Datasets No The paper is theoretical and does not use datasets for empirical evaluation. It assumes a Gaussian distribution as a theoretical model, but no concrete dataset is specified or made available.
Dataset Splits No The paper is theoretical and does not involve empirical evaluation or data splitting for training, validation, or testing.
Hardware Specification No The paper is theoretical and does not describe any experiments that would require specific hardware specifications.
Software Dependencies No The paper is theoretical and does not describe any experiments that would require specific software dependencies with version numbers.
Experiment Setup No The paper is theoretical and does not describe any experimental setup, hyperparameters, or training configurations.