Method, system and medium for mapping spiking neural network to brain-inspired computing platform

Abstract

The invention discloses a method, system and medium for mapping a pulse neural network to a brain-inspired computing platform. The method of the invention includes SNN simulation and analysis, and extracts the topological structure and neuron behavior of the trained SNN network to form an undirected SNN Graph and pulse behavior trajectory; divide the extracted undirected SNN graph into multiple SNN sub-blocks, and optimize the pulse traffic between SNN sub-blocks; use a heuristic search algorithm to find an optimal mapping scheme to divide the SNN sub-blocks Assign to neuron cores on the target brain-inspired platform, and minimize the average number of pulse hops communicated on the target platform. The present invention can divide a large SNN into multiple sub-blocks and map them to the NoC-based brain-inspired processor so that they can be executed efficiently, which can make the division scheme found at the time optimal, and can also make the cost of finding the mapping scheme low. Reduce the time consumption of searching and ensure the execution speed and efficient execution.

Description

technical field [0001] The invention relates to computing technology of a pulse neural network and a brain-inspired computing platform, in particular to a method, system and medium for mapping a pulse neural network to a brain-inspired computing platform. Background technique [0002] Spiking Neural Network (SNN for short) is the third-generation Artificial Neural Network (ANN for short) inspired by brain science. It is currently widely used in pattern recognition, image recognition and other fields. The neurons in the SNN will accumulate the membrane voltage after receiving the stimulation, and when the voltage exceeds the threshold, they will be excited to generate a pulse signal. SNN has strong biological characteristics, and at the same time requires low energy consumption when simulated on a brain-like platform. Most of the brain-inspired platforms that have been developed so far are based on the Network on Chip (NoC for short), through which the numerous neuron nuclei...

AI Technical Summary

Problems solved by technology

In order to obtain accurate evaluation results, it is usually necessary to use hardware or a hardware simulator, but this will bring a lot of time consumption, resulting in a significant decrease in the execution speed of the search algorithm

Images