Automatic Machine Learning Method for Hyper-parameter Search

Abstract

Abstract Automatic Machine Learning (AutoML) uses automated data-driven methods to realize the selection of hyper-parameters, neural network architectures, regularization methods, etc., making machine learning techniques easier to apply and reducing dependence on experienced human experts. And hyper-parameter search based on automatic machine learning is one of the current research hotspots in the industry and academia. We mainly introduce the hyper-parameter search framework based on automatic machine learning and the common hyper-parameter search strategies. Combined with specific data sets, the classification accuracy of the model under different hyper-parameter search strategies is compared to find the model parameter configuration that can maximize the classification accuracy. Compared with the experience-based parameter adjustment method, the hyper-parameter search based on automatic machine learning can reduce labor costs, improve training efficiency, and automatically construct a dedicated convolutional neural network to maximize the model effect.