Target algorithm fitting method based on neural network, terminal and application

Abstract

The invention provides a target algorithm fitting method based on a neural network. The method comprises the steps: acquiring a target algorithm capable of being approximated by the neural network; performing one-time iteration on the target algorithm to obtain data sets of different input and output variables; using the input variable as an independent variable, using the output variable as a dependent variable, and using a multivariate polynomial to fit the input and output variables of one iteration; determining a single hidden layer neural network structure of a multivariate polynomial ina fitting single iteration process; and repeating the iteration process, and connecting the iteration processes of each time in series to obtain the deep neural network which can finally fit the wholetarget algorithm. Meanwhile, the invention provides a deep neural network obtained based on the method, a channel capacity and energy distribution optimization method based on a WMMSE algorithm and aterminal used for executing the method. According to the method, the fitting problem of a complex algorithm is solved, and the structural design of the neural network and the selection of the numberof layers and neurons of the neural network can be practically guided.

Description

technical field [0001] The present invention relates to a method in the field of deep learning technology, in particular, to a neural network-based target algorithm fitting method, terminal and application, wherein the fitting method refers to a multivariate polynomial, which is a practical Deep Neural Network Architecture Construction Techniques for Algorithmic Approximation. Background technique [0002] Whether in communication or other fields, there are many optimization algorithms used to solve problems such as resource allocation and signal processing. However, for some more complex problems, artificially designed optimization algorithms often have considerable complexity. Between theoretical design and real-time processing caused a serious discrepancy. [0003] At present, deep learning is a very successful tool, and some works have made preliminary explorations and attempts on how to use neural networks to approximate optimization algorithms, and experiments in diff...

AI Technical Summary

Problems solved by technology

[0007] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a neural network-based target algorithm fitting method, terminal and application, and can be actually applied to guide the neural network construction of the neural network fitting algorithm, such as can be used for Solve the problem of maximizing channel capacity and energy allocation in the process of resource allocation in full-duplex channel systems

Images