Exploring the Versatility of Spiking Neural Networks: Applications Across Diverse Scenarios

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Artificial Neural Networks (ANNs) are crucial for learning algorithms.
• Spiking Neural Networks (SNNs), the third generation of ANNs, excel in pattern recognition and diagnostic medicine.
• SNNs offer lower energy consumption than traditional ANNs, especially on neuromorphic hardware.

Purpose of the Study:

• To explore the versatility and potential applications of Spiking Neural Networks.
• To demonstrate a dual-step methodology for tailoring SNNs to diverse tasks.
• To evaluate SNN performance in epileptic seizure detection and wine classification.

Main Methods:

• A rough SNN is designed based on dataset attributes.
• An extensive grid search algorithm refines the network by optimizing hyperparameters.
• The methodology is tested on binary and multi-classification tasks for seizure detection and wine classification.

Main Results:

• SNN models achieved significantly lower energy consumption than ANNs in binary classification tasks.
• Nearly 100% accuracy was reached in binary classification scenarios, such as epileptic seizure detection.
• An approximate 90% accuracy was achieved in multi-class classification tasks, indicating room for improvement.

Conclusions:

• The proposed dual-step methodology effectively tailors SNNs for diverse applications.
• SNNs demonstrate high efficiency and accuracy, particularly in binary classification tasks.
• Further improvements are possible for SNNs in multi-class classification scenarios.