Hyper-parameter optimization and model selection in machine learning
Machine learning is a powerful tool for data analysis and prediction but requires hyper-parameter tuning to show the best performance for a target problem. The term “hyper-parameter” indicates the parameter that should be set before the learning process begins. In practice, the users of machine learning algorithms have to select an appropriate algorithm and invest effort in finding good hyper-parameters such as regularization coefficient, kernel parameters, and learning rate. In the case of adopting deep neural networks, they also have to design the network architecture such as the numbers of units and layers, and the network connectivity. Manually optimizing such hyper-parameters requires trial and errors and considerable effort. Therefore, the hyper-parameter optimization is a major topic of automated machine learning (AutoML) which aims to automate the design of machine learning process.
The goal of the course is to understand the difficulty of the hyper-parameter optimization and introduce some techniques, e.g., grid search, Bayesian optimization, evolutionary computation. The course will provide the programming exercise of hyper-parameter optimization to gain the practical understanding. Further, the advanced techniques of hyper-parameter optimization, e.g., the neural network architecture search, are explained.
Hyper-parameter optimization, machine learning, black-box optimization, Bayesian optimization, automated machine learning.
Python (including numpy, matplotlib, pandas, and scikit-learn packages).
Basic knowledge of machine learning, basic of python programming, basic of mathematics (calculus, linear algebra, and statistics).
Dr. Shinichi Shirakawa
Lecturer at Yokohama National University, Japan
Shinichi Shirakawa is a lecturer at Faculty of Environment and Information Sciences, Yokohama National University, Japan. He received his Ph.D. degree in engineering from Yokohama National University in 2009. He worked at Fujitsu Laboratories Ltd. from 2010 to 2012 as a researcher. His research interests include machine learning, deep learning, evolutionary computation, computer vision, and so on. He won the best paper award of Genetic and Evolutionary Computation Conference (GECCO) 2017.
Fields of interests: Deep Learning, Black-box Optimization, Computer Vision.
Contacts: [email protected]