Cost-Efficient Online Hyperparameter Optimization

TL;DR

This paper introduces a cost-efficient online hyperparameter optimization method that adaptively queries validation data, significantly reducing training costs while achieving expert-level performance on image classification tasks.

Contribution

It proposes a novel Bayesian optimization algorithm with a costly feedback setting that improves efficiency in online hyperparameter tuning.

Findings

Achieves human expert-level hyperparameter tuning performance.

Reduces computational overhead compared to standard methods.

Effective on CIFAR-10 and ImageNet100 datasets.

Abstract

Recent work on hyperparameters optimization (HPO) has shown the possibility of training certain hyperparameters together with regular parameters. However, these online HPO algorithms still require running evaluation on a set of validation examples at each training step, steeply increasing the training cost. To decide when to query the validation loss, we model online HPO as a time-varying Bayesian optimization problem, on top of which we propose a novel \textit{costly feedback} setting to capture the concept of the query cost. Under this setting, standard algorithms are cost-inefficient as they evaluate on the validation set at every round. In contrast, the cost-efficient GP-UCB algorithm proposed in this paper queries the unknown function only when the model is less confident about current decisions. We evaluate our proposed algorithm by tuning hyperparameters online for VGG and ResNet on CIFAR-10 and ImageNet100. Our proposed online HPO algorithm reaches human expert-level performance within a single run of the experiment, while incurring only modest computational overhead compared to regular training.