Machine learning neural network model optimization method, system and equipment

Abstract

The invention provides a machine learning neural network model optimization method, system and equipment, and the method comprises the following steps: 1) building a non-convex loss function of a system, and generalizing the non-convex loss function into a non-convex nonlinear optimization problem; 2) solving a non-convex nonlinear optimization problem by using a group search variable parameter recurrent neural network; and 3) according to the parameters solved in the step 2, optimizing the parameters of the neural network in machine learning to minimize the loss function. According to the method, the thoughts of dual-time scale neurodynamics and particle swarm optimization are combined, and a novel group exploration neural dynamic network is obtained to solve the non-convex loss function optimization problem of a neural network in machine learning. The method has the capability of searching the global optimal solution in one iteration, the model solving efficiency is greatly improved, and the loss function is ensured to reach the optimal value, so that the optimized machine learning model can be obtained, and the precision in image recognition or image classification can be effectively improved through the machine learning model.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence planning optimization, and in particular relates to a machine learning neural network model optimization method and system. Background technique [0002] Machine learning is very important and widely used in the field of artificial intelligence, such as image recognition, image classification and so on. A key component in machine learning is loss function optimization, which is a function used to measure the gap between the output of a machine learning model and the actual output. By optimizing this function, the optimal machine learning model can be obtained. The loss function model is usually a non-convex nonlinear function model, but the optimization problem is usually difficult to solve. In the current loss function optimization technology, most of them use relaxation technology to relax the non-convex optimization into a convex optimization and then solve it. This method can...

AI Technical Summary

Problems solved by technology

This convex relaxation technique cannot obtain the optimal solution of non-convex optimization problems, and at the same time increases the computational difficulty

In the patent application document "Computer Vision Single Target Tracking Method Based on Optimal Variance Descent", Xing Weiwei et al. designed a deep neural network model based on non-convex optimization stochastic variance descent gradient, but the stochastic gradient model can only solve a local optimal solution. Unable to find the global optimal solution

Due to the development of artificial neural networks, many scholars use artificial recursive neural networks to optimize non-convex loss functions. The existing recurrent neural networks for solving non-convex problems can only find a certain global optimum through repeated iterations, which brings computational time. Shortcomings such as length and complexity make it impossible to quickly obtain the optimal network model

Images