Basic Enhancement Strategies When Using Bayesian Optimization for Hyperparameter Tuning of Deep Neural Networks

Compared to the traditional machine learning models, deep neural networks (DNN) are known to be highly sensitive to the choice of hyperparameters. While the required time and effort for manual tuning has been rapidly decreasing for the well developed and commonly used DNN architectures, undoubtedly DNN hyperparameter optimization will continue to be a major burden whenever a new DNN architecture needs to be designed, a new task needs to be solved, a new dataset needs to be addressed, or an existing DNN needs to be improved further. For hyperparameter optimization of general machine learning problems, numerous automated solutions have been developed where some of the most popular solutions are based on Bayesian Optimization (BO). In this work, we analyze four fundamental strategies for enhancing BO when it is used for DNN hyperparameter optimization. Specifically, diversification, early termination, parallelization, and cost function transformation are investigated. Based on the analysis, we provide a simple yet robust algorithm for DNN hyperparameter optimization - DEEP-BO (Diversified, Early-termination-Enabled, and Parallel Bayesian Optimization). When evaluated over six DNN benchmarks, DEEP-BO mostly outperformed well-known solutions including GP-Hedge, BOHB, and the speed-up variants that use Median Stopping Rule or Learning Curve Extrapolation. In fact, DEEP-BO consistently provided the top, or at least close to the top, performance over all the benchmark types that we have tested. This indicates that DEEP-BO is a robust solution compared to the existing solutions. The DEEP-BO code is publicly available at https://github.com/snu-adsl/DEEP-BO.