Working memory hardware implementation method based on reserve pool calculation

Abstract

The invention discloses a work memory hardware implementation method based on reserve pool calculation, and belongs to the technical field of artificial intelligence and life science. The method is implemented through the following system which comprises a reserve pool calculation module and a readout calculation module, and the reserve pool calculation module comprises a reserve pool input part,a reserve pool calculation part and a reserve pool output part; the readout calculation module comprises a nerve excitation type network, a nerve inhibition type network, an activation function part and a full connection layer part. According to the invention, a reserve pool calculation method and a working memory concept are combined, a reserve pool calculation network is built by utilizing a memory unit with a specific property, the reserve pool calculation network has a memory function for input information, can enable the past input information to be associated with current input information, so that the relation between neuron input stimulation can be better mastered; suppression and excitation of neuron stimulation are completed through the readout calculation module, the issuing characteristic of working memory is achieved, and hardware simulation of the working memory is well achieved.

Description

Technical field[0001]The invention relates to the technical fields of artificial intelligence and life sciences, in particular to a working memory hardware realization method based on reserve pool calculation.Background technique[0002]In 2001, Jaeger modified the traditional recurrent neural network, using the nonlinear Sigmoid function to simulate the nonlinear state of neurons, and called the improved network the echo state neural network; in the same year, Maass introduced the fluid state machine, which states The idea of ​​the machine network is the same as that of the echo state neural network, except that the basis of the fluid state machine is neural calculation, and the basis of the echo state neural network is machine learning. The reserve pool calculation is a general neuromorphic calculation method. State neural network and fluid state machine are developed. Compared with echo state neural network and fluid state machine, the reserve pool calculation algorithm is more con...

AI Technical Summary

Problems solved by technology

However, in the current field of life sciences, people are still at the stage of using MEMS probe technology, CMOS nano-electrodes and other technologies to obtain data inside and between biological neurons, and cannot directly imitate neurons through hardware according to neuron input stimulation. Therefore, how to simulate the working memory characteristics of neurons through software algorithms and hardware is an urgent problem to be solved.

Images