Spiking Neural Network Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pattern recognition systems, particularly in spiking neural networks, face challenges in efficiently comparing and segregating signals without explicit distance computation, especially in real-time applications like speech recognition and source separation, where training and supervision are required, and they struggle with noise, symmetry, and size changes.
Innovation Solution
A spiking neural network system with a global regulating unit and synchronized spiking neurons, using dynamic synaptic weights and excitatory/inhibitory connections to process patterns without explicit distance computation, allowing for unsupervised learning and efficient pattern recognition and segregation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional pattern recognition systems are used to compare and segregate signals, then pattern recognition capability is achieved, but computational burden increases due to explicit distance computation
Solution Approach 1:
The patent replaces conventional mechanical computation systems with a spiking neural network that uses biologically-inspired spike-based communication. The system substitutes traditional distance computation algorithms with neural spike propagation and temporal correlation mechanisms, where neurons naturally compute similarities through synchronized spiking patterns without explicit mathematical distance calculations.
Solution Approach 2:
The patent changes the fundamental parameter of signal representation from continuous amplitude values to discrete spike timing events. By transforming the representation parameter from analog to event-based temporal coding, the system eliminates the need for continuous distance computation and enables more efficient pattern comparison through temporal correlation of spike trains.
2Measurement precision
If training and supervision are used in spiking neural networks for source separation, then recognition accuracy improves, but system complexity and data requirements increase
Solution Approach 1:
The patent implements self-organizing maps that automatically learn and adapt to input patterns without external supervision or training data. The neural network performs unsupervised learning through competitive learning mechanisms where neurons self-organize their receptive fields based on incoming spike patterns, eliminating the need for labeled training datasets and complex training procedures.
Solution Approach 2:
The patent employs dynamic synaptic weights that automatically adjust based on temporal correlation of spike patterns. The synaptic connections are not static but evolve dynamically in response to incoming signals, allowing the network to adapt to different source separation scenarios without retraining, thereby reducing system complexity while maintaining accuracy.
3Extent of automation
If synchronized spiking neurons with dynamic synaptic weights are used, then unsupervised learning capability is achieved, but network complexity increases
Solution Approach 1:
The patent segments the neural network into distinct functional layers: an input layer that receives spike patterns, a hidden layer with self-organizing neurons that perform unsupervised learning, and an output layer that produces segmentation results. This segmentation allows each layer to have specialized, relatively simple functions while achieving complex unsupervised learning capability at the system level.
Data Source
AI summary
A spiking neural network has a layer of connected neurons exchanging signals. Each neuron is connected to at least one other neuron. A neuron is active if it spikes at least once during a time interval. Time-varying synaptic weights are computed between each neuron and at least one other neuron connected thereto. These weights are computed according to a number of active neurons that are connected to the neuron. The weights are also computed according to an activity of the spiking neural network during the time interval. Spiking of each neuron is synchronized according to a number of active neurons connected to the neuron and according to the weights. A pattern is submitted to the spiking neural network for generating sequences of spikes, which are modulated over time by the spiking synchronization. The pattern is characterized according to the sequences of spikes generated in the spiking neural network.