Average Case Column Subset Selection for Entrywise $\ell_1$-Norm Loss
This provides the first algorithm for (1+ε)-approximation in entrywise ℓ1-norm low-rank approximation, addressing a computational bottleneck in robust matrix approximation for data analysis applications.
The paper tackles the column subset selection problem for entrywise ℓ1-norm loss, showing that under realistic distributional assumptions (input matrix as a rank-k matrix plus i.i.d. noise with finite moment), a (1+ε)-approximation can be achieved with nearly linear time and poly(k/ε)+O(k log n) columns, while proving impossibility without a first moment.
We study the column subset selection problem with respect to the entrywise $\ell_1$-norm loss. It is known that in the worst case, to obtain a good rank-$k$ approximation to a matrix, one needs an arbitrarily large $n^{Ω(1)}$ number of columns to obtain a $(1+ε)$-approximation to the best entrywise $\ell_1$-norm low rank approximation of an $n \times n$ matrix. Nevertheless, we show that under certain minimal and realistic distributional settings, it is possible to obtain a $(1+ε)$-approximation with a nearly linear running time and poly$(k/ε)+O(k\log n)$ columns. Namely, we show that if the input matrix $A$ has the form $A = B + E$, where $B$ is an arbitrary rank-$k$ matrix, and $E$ is a matrix with i.i.d. entries drawn from any distribution $μ$ for which the $(1+γ)$-th moment exists, for an arbitrarily small constant $γ> 0$, then it is possible to obtain a $(1+ε)$-approximate column subset selection to the entrywise $\ell_1$-norm in nearly linear time. Conversely we show that if the first moment does not exist, then it is not possible to obtain a $(1+ε)$-approximate subset selection algorithm even if one chooses any $n^{o(1)}$ columns. This is the first algorithm of any kind for achieving a $(1+ε)$-approximation for entrywise $\ell_1$-norm loss low rank approximation.