Multi-task sparse Bayesian extreme learning machine regression method

Abstract

The invention provides a multi-task sparse Bayesian extreme learning machine regression method. The method comprises the steps of random feature extraction from an input layer to a hidden layer of a single hidden layer neural network, multi-task sparse modeling and posteriori estimation of an output layer weight, multi-task sparse Bayesian extreme learning machine parameter and hyper-parameter rapid optimization estimation and the like. According to the method, a hierarchical Bayesian model is adopted to carry out multi-task sparse solution on an output layer weight of an extreme learning machine; on the premise that the precision is guaranteed, redundant hidden layer neurons of the extreme learning machine are cut, a more compact neural network is obtained, the over-fitting phenomenon ofthe extreme learning machine is effectively avoided, and the number of the hidden layer neurons does not need to be determined in advance. From the perspective of sparse Bayesian learning, the singlehidden layer neural network at the front end can enable the sparse Bayesian learning method to be applied to nonlinear problems.

Description

technical field [0001] The invention belongs to the technical field of machine learning and civil engineering, and in particular relates to a multi-task sparse Bayesian extreme learning machine regression method, which is applicable to statistically related data that are not from the same regression task. Background technique [0002] In today's era of increasingly large amounts of data, the concept of artificial intelligence has received more and more attention. As the core part of artificial intelligence, machine learning algorithm is the part that gives machines the ability to think like a human. Obtaining the implicit functional relationship in data through machine learning algorithms is more and more widely used in our daily life, solving many problems that could not be solved before. For example, in the field of geotechnical engineering, for cohesive soils, the correlation among the five parameters of fluidity index, vertical effective stress, pre-consolidation stress...

AI Technical Summary

Problems solved by technology

[0004] However, there are the following very obvious deficiencies in the extreme learning machine algorithm: 1. The extreme learning machine solves the weight vector of the output layer through the least square method, which is very easy to produce overfitting

2. The performance of the extreme learning machine is heavily dependent on the number of neurons in the hidden layer

[0008] At present, there are not many examples of using extreme learning machine algorithms for multi-task learning, and there are few related theoretical researches.

Images