Improved Regret Guarantees for Online Mirror Descent using a Portfolio of Mirror Maps

TL;DR

This paper demonstrates that selecting appropriate mirror maps, especially block norm-based ones, can significantly improve regret bounds in online convex optimization, particularly for sparse loss functions, through adaptive algorithms.

Contribution

It introduces a family of mirror maps based on block norms that adapt better to sparsity, and proposes a meta-algorithm to dynamically select geometries, achieving polynomial regret improvements.

Findings

Block norm-based mirror maps outperform traditional $L_1$ and $L_2$ geometries.

Naive switching between geometries can cause linear regret.

Meta-algorithm effectively adapts to unknown sparsity levels, improving regret guarantees.

Abstract

OMD and its variants give a flexible framework for OCO where the performance depends crucially on the choice of the mirror map. While the geometries underlying OPGD and OEG, both special cases of OMD, are well understood, it remains a challenging open question on how to construct an optimal mirror map for any given constrained set and a general family of loss functions, e.g., sparse losses. Motivated by parameterizing a near-optimal set of mirror maps, we consider a simpler question: is it even possible to obtain polynomial gains in regret by using mirror maps for geometries that interpolate between $L_1$ and $L_2$, which may not be possible by restricting to only OEG ($L_1$) or OPGD ($L_2$). Our main result answers this question positively. We show that mirror maps based on block norms adapt better to the sparsity of loss functions, compared to previous $L_p$ (for $p \in [1, 2]$) interpolations. In particular, we construct a family of online convex optimization instances in $\mathbb{R}^d$, where block norm-based mirror maps achieve a provable polynomial (in $d$) improvement in regret over OEG and OPGD for sparse loss functions. We then turn to the setting in which the sparsity level of the loss functions is unknown. In this case, the choice of geometry itself becomes an online decision problem. We first show that naively switching between OEG and OPGD can incur linear regret, highlighting the intrinsic difficulty of geometry selection. To overcome this issue, we propose a meta-algorithm based on multiplicative weights that dynamically selects among a family of uniform block norms. We show that this approach effectively tunes OMD to the sparsity of the losses, yielding adaptive regret guarantees. Overall, our results demonstrate that online mirror-map selection can significantly enhance the ability of OMD to exploit sparsity in online convex optimization.