Spiking Neural Network Synaptic Weight Training via Segmented Plasticity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current synaptic weight training methods for spiking neural networks are inefficient in terms of resource usage and training speed, and existing target identification methods are not optimized for rapid and accurate identification.
Innovation Solution
A method that combines back propagation and synaptic plasticity rules to quickly train and update synaptic weights in spiking neural networks, enabling rapid target identification by generating spike sequences from images and inputting them into a trained network for efficient processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional back propagation method is used for training spiking neural networks, then training accuracy can be achieved, but training speed is slow and resource occupation is high
Solution Approach 1:
The patent segments the training process into two distinct phases: a first training phase using back propagation to obtain initial synaptic weights, and a second training phase using spike timing dependent plasticity to refine weights. This segmentation allows each phase to specialize in different aspects of learning, improving overall training efficiency while reducing resource requirements compared to using only traditional back propagation.
Solution Approach 2:
The patent changes the training parameters by introducing spike timing dependent plasticity mechanisms that modify synaptic weights based on the relative timing of pre-synaptic and post-synaptic spikes. This parameter change enables the network to learn temporal patterns more efficiently, accelerating training convergence and reducing computational resource occupation.
2Speed
If traditional neural networks are used for target identification, then identification accuracy can be achieved, but identification speed is slow
Solution Approach 1:
The patent implements dynamic spike timing dependent plasticity mechanisms that allow synaptic weights to adapt in real-time based on spike timing relationships. This dynamic adjustment enables the spiking neural network to rapidly process and identify targets while maintaining high accuracy, as the weights continuously optimize based on temporal patterns in the input data.
Solution Approach 2:
The patent replaces traditional continuous-valued neural network mechanisms with discrete spike-based mechanisms. By using spike events rather than continuous activations, the system achieves faster processing speeds while the spike timing dependent plasticity rules ensure that identification accuracy is maintained through temporal coding information.
Data Source
AI summary
The present disclosure relates to a synaptic weight training method, a target identification method, an electronic device and a storage medium. The synaptic weight training method includes: inputting spike signals corresponding to training samples into a spiking neural network to be trained; learning a target synaptic weight in the spiking neural network with a back propagation rule to obtain a first branch weight; and learning the target synaptic weight with a synaptic plasticity rule to obtain a second branch weight; and updating the target synaptic weight according to the first branch weight and the second branch weight.