A random neural network hardware realization device

A stochastic neural network and hardware implementation technology, applied in biological neural network models, physical realization, etc., can solve the problems of complex models and low accuracy of neural networks

Active Publication Date: 2019-01-11
SHANGHAI UNIV
View PDF5 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0008] The object of the present invention is to provide a random neural network hardware implementation device with high precision, good real-time performance and fault tolerance, aiming at the defects of low precision and complex models of the above-mentioned neural network.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A random neural network hardware realization device
  • A random neural network hardware realization device
  • A random neural network hardware realization device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0038] refer to figure 1 , image 3 , Figure 5 and Figure 7 , a schematic diagram of the system design of a random neural network hardware implementation device in a preferred embodiment of the present invention, which includes a three-layer structure—an input layer, a hidden layer and an output layer: the input layer consists of m input neurons I Each input neuron I includes a random number converter A, and the input vector (1) outputs random data sequence one (2) after passing through the random number converter A; the hidden layer consists of s hidden neurons J, Each hidden neuron J includes a random number converter B, a random function generator (41) and a deterministic number converter C. After passing through the random number converter B, the parameter code stream one (11) is combined with the random data sequence one. (2) After passing through the random function generator (41) together, the random data sequence two (13) is obtained, and the random data sequence ...

Embodiment 2

[0040] refer to figure 2 It is a schematic structural diagram of a hardware implementation device of a random neural network in a preferred embodiment of the present invention. The neural network includes m input neurons I, s hidden neurons J, and n output neurons K. The number of neurons in each layer is set according to different applications. There is an optimal number of input layer nodes m and The number of hidden layer nodes is s, which makes the network structure have high calculation accuracy. Input neuron I accepts input vector (1), passes through random number converter A, and outputs random data sequence one (2); implicit neuron J accepts random data sequence one (2), parameter code stream one (11), parameter Digital stream one (11) obtains random code stream sequence (12), random code stream sequence (12) and random data sequence one (2) through random number converter B and input it to random function generator (41) to obtain random data sequence Two (13), the ...

Embodiment 3

[0043] This embodiment is basically the same as the first embodiment, and the special features are as follows:

[0044] The hardware implementation device of a random neural network further includes a parameter code stream 1 (11) and a parameter code stream 2 (21), both of which are stored in a non-volatile memory. Among them, the parameter code stream 1 (11) is involved in calculating the output value of the hidden neuron J (13), the parameter code stream 2 (21) is involved in calculating the output value of the output neuron K (23), and the parameter code stream 1 (11) Can be random or non-random. When the parameter code stream one is a non-random sequence, the hardware implementation device for a random neural network further includes a random number converter B, which is used to convert the parameter code stream one (11) into a random sequence. The second parameter stream (21) can be a random sequence or a non-random sequence. When the second parameter code stream ( 21 )...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention relates to a random neural network hardware realization apparatus. The apparatus comprises an input layer, a hidden layer and an output layer, wherein the input layer consists of m input neurons I, each input neuron I comprises a random number converter A, and after an input vector passes through the random number converter A, a random data sequence I is output; the hidden layer consists of s hidden neurons J, each hidden neuron J comprises a random number converter B, a random function generator and a definite number converter C, a parameter code stream passes through the random number converter B, is aligned to the random data sequence and passes through the random function generator, so a random data sequence II is obtained, and the random number sequence II passes through the definite number converter C and a definite number I is output; the output layer consists of n output neurons K, each output neuron K comprises a definite number converter D and a linear function processor, and a parameter code stream II passes through the definite number converter D, is aligned to the definite number and passes through the linear function processor, so an object vector is output. According to the apparatus, hardware logic and wiring resources can be greatly reduced, circuit cost and power consumption can be reduced, the network operation precision is high, and the fitting capability of training samples is enhanced.

Description

technical field [0001] The invention relates to a hardware implementation device of a random neural network. The so-called random neural network refers to a network structure in which the input layer and the hidden layer use random numbers to transmit and process data, and the output layer uses deterministic numbers to output. Background technique [0002] Artificial Neural Network (Neural Network for short) is a complex network formed by interconnecting a large number of simple neurons by imitating the structure and function of the neural network of the brain from the perspective of information processing. Each neuron accepts input from a large number of other neurons and produces outputs through a parallel network, affecting other neurons. The mutual restriction and mutual influence between the networks realizes the nonlinear mapping from the input state to the output state space. The artificial neural network can obtain the weight and structure of the network through tr...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Patents(China)
IPC IPC(8): G06N3/06
CPCG06N3/06
Inventor 季渊王雪纯陈文栋冉峰满丽萍
Owner SHANGHAI UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap