A neuromorphic network for generic multivariate data classification.

Area of Science:

• Computational Neuroscience
• Neuromorphic Engineering
• Machine Learning

Background:

• Computational neuroscience identifies brain computational principles.
• Neuromorphic hardware enables efficient neural network implementation.
• Implementing functional brain algorithms on hardware is a key challenge.

Purpose of the Study:

• To construct a spiking neural network for multivariate data classification, inspired by insect olfactory systems.
• To implement and evaluate a functional brain algorithm on neuromorphic hardware.

Main Methods:

• Developed a spiking neural network using virtual receptors (VRs) to convert data into spike trains.
• Implemented lateral inhibition and winner-take-all circuits for supervised learning.
• Ran classification and inhibition stages on accelerated neuromorphic hardware.

Main Results:

• The spiking neural network achieved classification performance on par with a Naïve Bayes classifier on real-world datasets.
• Network decision-making occurred within 100 ms, matching insect nervous system response times.
• The network demonstrated robustness to hardware variability through population coding.

Conclusions:

• The study provides a proof of principle for implementing functional spiking neural networks on configurable neuromorphic hardware.
• This approach is suitable for real-world computing problems requiring fast and robust classification.
• The findings bridge computational neuroscience principles with practical neuromorphic applications.