19
3

Nonparametric Matrix Estimation with One-Sided Covariates

Abstract

Consider the task of matrix estimation in which a dataset XRn×mX \in \mathbb{R}^{n\times m} is observed with sparsity pp, and we would like to estimate E[X]\mathbb{E}[X], where E[Xui]=f(αu,βi)\mathbb{E}[X_{ui}] = f(\alpha_u, \beta_i) for some Holder smooth function ff. We consider the setting where the row covariates α\alpha are unobserved yet the column covariates β\beta are observed. We provide an algorithm and accompanying analysis which shows that our algorithm improves upon naively estimating each row separately when the number of rows is not too small. Furthermore when the matrix is moderately proportioned, our algorithm achieves the minimax optimal nonparametric rate of an oracle algorithm that knows the row covariates. In simulated experiments we show our algorithm outperforms other baselines in low data regimes.

View on arXiv
Comments on this paper