报告题目 (Title)：An Introduction to Hyperparameter Optimization(超参数优化问题简介)
报告人 (Speaker)：方慧副教授（上海财经大学）
报告时间 (Time)：2021年12月14日(周二) 16:00 - 17:00
报告地点 (Place)：G507
邀请人(Inviter)：余长君
报告摘要：Machine learning (ML) has been widely exploited in both academia and industry. Building an effective machine learning model is a time-consuming process that involves obtaining an optimal model architecture with fine-tuned hyperparameters. Besides, recent interest in complex ML models with a relatively large volume of hyperparameters (e.g., autoML and deep learning methods) has resulted in an increasing volume of studies on hypeparameter optimization (HPO).
In this talk, Iwill first formally define the HPO problem, and give an overview of existing wok in this field of research. Secondly, three types of HPO methods, i.e., sampling-based, model-based and gradient-based, are elaborated. Finally, I will conclude the talk by summarizing challenging issues on the topic.