PHD Defense: Hongsheng Liu
Hybrid Bayesian Optimization with DIRECT
Machine learning has been a topic in academia and industry for decades. The performance of machine learning models heavily relies on the hyperparameter selections. We work on the hyperparameter tuning optimization for machine learning models. We propose a hybrid Bayesian optimization method using the DIRECT algorithm which provides a new strategy to tradeoff between exploitation and exploration. Convergence results are established under suitable assumptions and numerical experiments on benchmark problems show the efficiency of the new algorithm.