A new supervised learning algorithm for multiple spiking neural networks with application in epilepsy and seizure...

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Traditional Spiking Neural Networks (SNNs) have limitations in efficiency and complexity.
• Previous work by the authors improved single-spiking SNN efficiency significantly.

Purpose of the Study:

• To introduce a novel Multi-Spiking Neural Network (MuSpiNN) model.
• To develop a supervised learning algorithm, Multi-SpikeProp, for training MuSpiNN.
• To evaluate the performance of MuSpiNN and Multi-SpikeProp on classification tasks.

Main Methods:

• MuSpiNN transmits information via multiple spikes through multiple synapses.
• Multi-SpikeProp is a supervised learning algorithm for MuSpiNN training.
• A modular architecture was used for complex classification problems (iris, EEG).

Main Results:

• MuSpiNN learned the XOR problem with fewer synapses than single-spiking SNNs.
• For EEG classification, MuSpiNN achieved 90.7%-94.8% accuracy, surpassing single-spiking SNNs (82%).
• The model leverages optimized heuristic rules and parameters from prior research.

Conclusions:

• MuSpiNN and Multi-SpikeProp demonstrate improved efficiency and accuracy for complex neural network tasks.
• The multi-spiking approach offers advantages in synapse reduction and classification performance.
• This model shows significant potential for applications like EEG-based seizure detection.