Fraction linear neural network model

Abstract

The invention advances a fractional linear neural network model, comprising m layers of neural cells, where the input layer has 1(x) neural cells, the ith layer has i(x) neural cells and the output layer has m(x) neural cells; the neural cells of all layers are forward interlinked, and the neural cells in the same layer are not interlinked, the input sum of the ith neural cell of the kth layer is written into Iik< / SUP, and the output written into oik, the weight value of connection of the ith neural cell of the (k-1)th hidden layer to the jth neural cell of the kth layer is written into wiú¼jk-1ú¼k, the output function of the jth neural cell of the kth layer is written into fjkú¼and thus the operating characteristics of the ith neutral cell of the (k-1)th hidden layer and the jth neutral cell of the kth layer are written (see top fig.). The invention solves the difficult problem that there are no modeling and policy making tools for nonlinear continuous mapping in infinite regions.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, and proposes a fractional linear neural network model. technical background [0002] Topologically, the artificial neural network can be regarded as a directed graph with processing units, also called neurons, as nodes and connected by weighted directed arcs. The processing unit is the simulation of physiological neurons, and the directed arc is the simulation of the "axon-dendrite" pair; the weight of the directed arc marks the strength of the mutual influence between the two processing units. The interconnection strength matrix formed by integrating all directed arcs corresponds to the long-term memory of information in the human brain. The processing unit uses a nonlinear function to realize the nonlinear mapping between the input and output of the unit, and its immediate active value corresponds to the short-term memory of the information in the human brain. Set of processin...

AI Technical Summary

Problems solved by technology

[0013] 2. It cannot be used as a pattern classifier in the infinite neighborhood;

[0014] 3. The existing artificial neural network cannot infinitely approach the nonlinear bounded continuous mapping on the unbounded area, and cannot provide modeling tools on the infinite domain for engineering technical problems

[0015] However, in many practical engineering fields, such as aerospace, etc., people often encounter modeling problems of continuous maps f on unbounded closed subsets, classification on unbounded regions, decision-making problems

Images