Abstract
Traditionally, robust statistics has focused on designing estimators tolerantto a minority of contaminated data. Robust list-decodable learning focuses onthe more challenging regime where only a minority $\frac 1 k$ fraction of thedataset is drawn from the distribution of interest, and no assumptions are madeon the remaining data. We study the fundamental task of list-decodable meanestimation in high dimensions. Our main result is a new list-decodable meanestimation algorithm for bounded covariance distributions with optimal samplecomplexity and error rate, running in nearly-PCA time. Assuming the groundtruth distribution on $\mathbb{R}^d$ has bounded covariance, our algorithmoutputs a list of $O(k)$ candidate means, one of which is within distance$O(\sqrt{k})$ from the truth. Our algorithm runs in time $\widetilde{O}(ndk)$for all $k = O(\sqrt{d}) \cup \Omega(d)$, where $n$ is the size of the dataset.We also show that a variant of our algorithm has runtime $\widetilde{O}(ndk)$for all $k$, at the expense of an $O(\sqrt{\log k})$ factor in the recoveryguarantee. This runtime matches up to logarithmic factors the cost ofperforming a single $k$-PCA on the data, which is a natural bottleneck of knownalgorithms for (very) special cases of our problem, such as clusteringwell-separated mixtures. Prior to our work, the fastest list-decodable meanestimation algorithms had runtimes $\widetilde{O}(n^2 d k^2)$ and$\widetilde{O}(nd k^{\ge 6})$. Our approach builds on a novel soft downweighting method, $\mathsf{SIFT}$,which is arguably the simplest known polynomial-time mean estimation techniquein the list-decodable learning setting. To develop our fast algorithms, weboost the computational cost of $\mathsf{SIFT}$ via a careful "win-win-win"analysis of an approximate Ky Fan matrix multiplicative weights procedure wedevelop, which we believe may be of independent interest.