Meta-strategy for Learning Tuning Parameters with Guarantees
This work addresses the practical difficulty of setting tuning parameters for online learning algorithms, which is a problem for practitioners using these methods. The improvements are incremental.
This paper proposes a meta-strategy to learn tuning parameters for online learning methods like OGA and EWA from past tasks. The strategy minimizes a regret bound, enabling the learning of initialization and step size in OGA, and prior or learning rate in EWA, with theoretical guarantees.
Online learning methods, like the online gradient algorithm (OGA) and exponentially weighted aggregation (EWA), often depend on tuning parameters that are difficult to set in practice. We consider an online meta-learning scenario, and we propose a meta-strategy to learn these parameters from past tasks. Our strategy is based on the minimization of a regret bound. It allows to learn the initialization and the step size in OGA with guarantees. It also allows to learn the prior or the learning rate in EWA. We provide a regret analysis of the strategy. It allows to identify settings where meta-learning indeed improves on learning each task in isolation.