16
74

List-Decodable Linear Regression

Abstract

We give the first polynomial-time algorithm for robust regression in the list-decodable setting where an adversary can corrupt a greater than 1/21/2 fraction of examples. For any α<1\alpha < 1, our algorithm takes as input a sample {(xi,yi)}in\{(x_i,y_i)\}_{i \leq n} of nn linear equations where αn\alpha n of the equations satisfy yi=xi,+ζy_i = \langle x_i,\ell^*\rangle +\zeta for some small noise ζ\zeta and (1α)n(1-\alpha)n of the equations are {\em arbitrarily} chosen. It outputs a list LL of size O(1/α)O(1/\alpha) - a fixed constant - that contains an \ell that is close to \ell^*. Our algorithm succeeds whenever the inliers are chosen from a \emph{certifiably} anti-concentrated distribution DD. In particular, this gives a (d/α)O(1/α8)(d/\alpha)^{O(1/\alpha^8)} time algorithm to find a O(1/α)O(1/\alpha) size list when the inlier distribution is standard Gaussian. For discrete product distributions that are anti-concentrated only in \emph{regular} directions, we give an algorithm that achieves similar guarantee under the promise that \ell^* has all coordinates of the same magnitude. To complement our result, we prove that the anti-concentration assumption on the inliers is information-theoretically necessary. Our algorithm is based on a new framework for list-decodable learning that strengthens the `identifiability to algorithms' paradigm based on the sum-of-squares method. In an independent and concurrent work, Raghavendra and Yau also used the Sum-of-Squares method to give a similar result for list-decodable regression.

View on arXiv
Comments on this paper