Neuromorphic Spiking Neural Network Multi-Neuron Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing spiking neural networks for pattern recognition are not feasible for low-SWaP neuromorphic hardware due to restrictive parameter requirements and limited application to neuromorphic hardware, as they allow only one excitatory neuron to spike at a time.
Innovation Solution
A system that trains a spiking neural network on neuromorphic hardware by generating spike trains for each excitatory neuron across multiple training patterns, allowing multiple excitatory neurons to fire simultaneously, and uses normalized spiking rate distributions to classify unlabeled input patterns, enabling efficient pattern recognition on low-SWaP devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a spiking neural network allows only one excitatory neuron to spike per time (as in Diehl and Cook's network), then the network can be implemented on neuromorphic hardware, but the pattern recognition capability and accuracy are limited
Solution Approach 1:
The patent changes the spiking parameter from single-neuron-spiking to multi-neuron-spiking mode. By allowing multiple excitatory neurons to spike simultaneously and adjusting the spiking rate distribution, the network achieves higher pattern recognition accuracy while remaining compatible with neuromorphic hardware constraints
Solution Approach 2:
The patent introduces dynamic spiking rate distribution where different excitatory neurons can have different spiking rates based on their contribution to pattern recognition. This dynamic parameter adjustment enables the network to optimize its performance for different input patterns while maintaining hardware feasibility
2Adaptability or versatility
If the spiking neural network uses restrictive parameter sets (as in Diehl and Cook's approach), then it can be applied to neuromorphic hardware, but the adaptability and versatility are reduced
Solution Approach 1:
The patent modifies key parameters including allowing multiple excitatory neurons to spike simultaneously, adjusting spiking rate distributions, and normalizing spike rates. These parameter changes enhance the network's adaptability to various input patterns while maintaining compatibility with low-SWaP neuromorphic hardware through careful parameter selection
Solution Approach 2:
The patent segments the neural network into distinct functional components with specific parameter ranges: input layer neurons, excitatory neurons with adjustable spiking rates, and inhibitory neurons. This segmentation allows each component to operate within hardware-friendly parameter constraints while collectively achieving high adaptability
3Reliability
If multiple excitatory neurons fire at the same time, then the pattern recognition capability improves, but the computational complexity and resource requirements increase
Solution Approach 1:
The patent implements continuous spiking activity where multiple excitatory neurons can fire simultaneously based on their normalized spiking rate distributions. This continuous multi-neuron spiking provides richer pattern recognition information while the normalization process ensures energy consumption remains within acceptable bounds for neuromorphic hardware
Data Source
AI summary
Described is a system for pattern recognition designed for neuromorphic hardware. The system generates a spike train of neuron spikes for training patterns with each excitatory neuron in an excitatory layer, where each training pattern belongs to a pattern class. A spiking rate distribution of excitatory neurons is generated for each pattern class. Each spiking rate distribution of excitatory neurons is normalized, and a class template is generated for each pattern class from the normalized spiking rate distributions. An unlabeled input pattern is classified using the class templates. A mechanical component of an autonomous device can be controlled based on classification of the unlabeled input pattern.