An efficient agnostic learner for halfspaces over subgaussian isotropic log-concave distributions returns a classifier with error opt+O(ε) and boundary volume O(r+ε) in time d^{~O(1/ε^2)}.
Bounded Independence Fools Halfspaces
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We show that any distribution on {-1,1}^n that is k-wise independent fools any halfspace h with error \eps for k = O(\log^2(1/\eps) /\eps^2). Up to logarithmic factors, our result matches a lower bound by Benjamini, Gurel-Gurevich, and Peled (2007) showing that k = \Omega(1/(\eps^2 \cdot \log(1/\eps))). Using standard constructions of k-wise independent distributions, we obtain the first explicit pseudorandom generators G: {-1,1}^s --> {-1,1}^n that fool halfspaces. Specifically, we fool halfspaces with error eps and seed length s = k \log n = O(\log n \cdot \log^2(1/\eps) /\eps^2). Our approach combines classical tools from real approximation theory with structural results on halfspaces by Servedio (Computational Complexity 2007).
fields
cs.DS 1years
2025 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Robust learning of halfspaces under log-concave marginals
An efficient agnostic learner for halfspaces over subgaussian isotropic log-concave distributions returns a classifier with error opt+O(ε) and boundary volume O(r+ε) in time d^{~O(1/ε^2)}.