Evolving spatio-temporal data machines based on the NeuCube neuromorphic framework: Design methodology and selected...

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Data Science

Background:

• Traditional systems struggle with large, fast spatio/spectro temporal data.
• Need for adaptive, interpretable machine learning models.

Purpose of the Study:

• Introduce evolving spatio-temporal data machines (eSTDM) based on neuromorphic principles.
• Present the NeuCube framework and methodology for eSTDM development.
• Demonstrate eSTDM applications in pattern recognition and early event prediction.

Main Methods:

• Utilizing spiking neural networks (SNN) as core processing modules.
• Developing multi-modular computer systems capable of incremental learning.
• Implementing the NeuCube framework in MATLAB, Java, and PyNN for neuromorphic hardware compatibility.

Main Results:

• eSTDM demonstrate continuous learning, adaptation, and interpretability.
• Spiking neural networks enable early event prediction.
• Successful application examples on diverse datasets (EEG, fMRI, seismic, climate, etc.).

Conclusions:

• eSTDM represent a novel approach for processing complex spatio-temporal data.
• The NeuCube framework facilitates eSTDM implementation and application.
• Future work includes neurogenetic extensions and new application domains.