A Bayesian optimization method based on a sequential Monte Carlo method

Abstract

The invention discloses a Bayesian optimization method based on a sequential Monte Carlo method, and the method comprises the following steps: S1, building an objective function, and obtaining the posterior distribution of the objective function; S2, adding a noise signal, and calculating an edge Stuent-t distribution of the objective function; and S3, optimizing the objective function through a sequential Monte Carlo approximation method. Weight distribution is calculated through a sequential Monte Carlo method; the improved sequential Monte Carlo method is expanded to global optimization through Student-t process regression instead of Gaussian process regression, maximum value distribution can be more effectively obtained under the condition that the number of samples is small, and higher exploration capacity and abnormal value adaptability are achieved.

Description

technical field

[0001] The invention belongs to the technical field of global optimization, and in particular relates to a Bayesian optimization method based on a sequential Monte Carlo method. Background technique

[0002] Many problems in science and engineering can be described as finding the minimum or maximum of an unknown function that is difficult to estimate. Bayesian optimization is a widely used probabilistic method for solving this problem. In order to explain the characteristics of unknown objective functions, Gaussian processes are particularly suitable for the interpretation of model predictions and provide a reasonable framework for learning and model selection. Due to its advantages in modulation and learning, it has become the most popular nonparametric kernel-based global optimization method in machine learning. It is widely used in many applications or research areas, for example, in reinforcement learning, big data, wireless sensor networks, and many ot...

Images