Online Sparse Linear Regression

TL;DR

This paper studies online sparse linear regression, providing an inefficient algorithm with sublinear regret and proving computational hardness for polynomial-time algorithms to achieve better regret bounds, thus resolving an open problem.

Contribution

It introduces an inefficient algorithm with sublinear regret and establishes a computational hardness barrier for polynomial-time algorithms in online sparse linear regression.

Findings

Inefficient algorithm achieves $ ilde{O}( oot{T} ull)$ regret.

No polynomial-time algorithm can achieve $O(T^{1- ext{delta}})$ regret unless NP $ extsubseteq$ BPP.

Hardness result holds even with access to more features than the sparse regressor.

Abstract

We consider the online sparse linear regression problem, which is the problem of sequentially making predictions observing only a limited number of features in each round, to minimize regret with respect to the best sparse linear regressor, where prediction accuracy is measured by square loss. We give an inefficient algorithm that obtains regret bounded by $\tilde{O}(\sqrt{T})$ after $T$ prediction rounds. We complement this result by showing that no algorithm running in polynomial time per iteration can achieve regret bounded by $O(T^{1-\delta})$ for any constant $\delta > 0$ unless $\text{NP} \subseteq \text{BPP}$. This computational hardness result resolves an open problem presented in COLT 2014 (Kale, 2014) and also posed by Zolghadr et al. (2013). This hardness result holds even if the algorithm is allowed to access more features than the best sparse linear regressor up to a logarithmic factor in the dimension.