23
3

Robust Methods for High-Dimensional Linear Learning

Abstract

We propose statistically robust and computationally efficient linear learning methods in the high-dimensional batch setting, where the number of features dd may exceed the sample size nn. We employ, in a generic learning setting, two algorithms depending on whether the considered loss function is gradient-Lipschitz or not. Then, we instantiate our framework on several applications including vanilla sparse, group-sparse and low-rank matrix recovery. This leads, for each application, to efficient and robust learning algorithms, that reach near-optimal estimation rates under heavy-tailed distributions and the presence of outliers. For vanilla ss-sparsity, we are able to reach the slog(d)/ns\log (d)/n rate under heavy-tails and η\eta-corruption, at a computational cost comparable to that of non-robust analogs. We provide an efficient implementation of our algorithms in an open-source Python\mathtt{Python} library called linlearn\mathtt{linlearn}, by means of which we carry out numerical experiments which confirm our theoretical findings together with a comparison to other recent approaches proposed in the literature.

View on arXiv
Comments on this paper