NADOL: Neuromorphic Architecture for Spike-Driven Online Learning by Dendrites

Area of Science:

• Neuromorphic Engineering
• Computational Neuroscience
• Artificial Intelligence

Background:

• Spike-driven learning in neuromorphic computing faces challenges with on-line learning capabilities.
• Neuronal dendrites offer a promising avenue for enhancing learning by introducing dendritic processing as a hyperparameter.
• Neuromorphic computing is crucial for advancing spike-based machine intelligence and neural learning systems.

Purpose of the Study:

• To introduce a novel neuromorphic architecture with dendritic on-line learning (NADOL) for brain-inspired intelligence on embedded hardware.
• To address the open challenge of on-line learning in neuromorphic models.
• To enhance learning efficiency, convergence speed, and reduce power consumption.

Main Methods:

• Developed the Neuromorphic Architecture with Dendritic On-line Learning (NADOL).
• Utilized distributed processing with spiking neural networks.
• Employed a piecewise linear approximation to reduce computational cost.
• Analyzed the impact of neuron count, dendritic segregation, feedback, and connection sparseness on learning.

Main Results:

• NADOL demonstrates remarkable learning capabilities, surpassing existing solutions.
• Achieved superior performance compared to GPU platforms in dendritic learning.
• Showcased enhanced learning efficiency and convergence speed.
• Reduced power consumption through distributed processing.

Conclusions:

• NADOL presents an efficient methodology for brain-inspired intelligence on embedded hardware.
• The architecture effectively integrates biologically plausible dendritic processing for improved on-line learning.
• NADOL signifies a step towards bridging artificial intelligence and neuroscience, with applications in IoT, robotics, and brain-machine interfaces.