A 3D Convolutional Neural Network Implementation Method Based on Memristor

Abstract

The invention discloses a method for implementing a three-dimensional convolutional neural network based on a memristor, which includes preparing a three-dimensional convolution to be trained consisting of a first convolution+pooling layer, a second convolution+pooling layer, and a fully connected layer Neural network, and input the original video information to obtain the output value of the fully connected layer; according to the deviation between the output value of the fully connected layer of the three-dimensional convolutional neural network to be trained and the standard information, use the backpropagation function to train the three-dimensional convolutional neural network to be trained ; When the set number of training times is reached, it is determined whether the training accuracy is up to the standard, and if it is not up to the standard, retraining is carried out until the accuracy reaches the standard; steps such as obtaining the required three-dimensional convolutional neural network. Its remarkable effect is: reducing the impact on precision, preparation difficulty and time on the hardware; it is convenient for hardware construction and realization, and at the same time, for the first time, the difficulty of information that can be processed by the neural network based on the memristor is raised to the video level.

Description

technical field [0001] The invention relates to the technical fields of artificial intelligence and material science, in particular to a method for realizing a three-dimensional convolutional neural network based on a memristor. Background technique [0002] Memristor is a dimension with resistance, as the fourth basic element following the resistor representing the relationship between voltage and current, the capacitor representing the relationship between charge and voltage, and the inductor representing the relationship between magnetic flux and current. A circuit device that represents the relationship between magnetic flux and charge (dρ=Mdq). The resistance value of the memristor will change with the amount of current flowing through the memristor. Even if the current stops, its resistance value will still stop at the value before the current stops, unless the current flows through the memristor again. For memristors, if the current flows in the forward direction, th...

AI Technical Summary

Problems solved by technology

[0005] However, most of the neural networks implemented based on memristors are the most basic fully connected layers or simple convolutional layers, which limits the use of memristor characteristics, and the existing research is limited to the recognition of simple signal and image models.

Images