A neuromorphic architecture for object recognition and motion anticipation using burst-STDP

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Neuromorphic Engineering

Background:

• Artificial spiking neurons offer a minimal framework for in silico deployment.
• Spike-Timing Dependent Plasticity (STDP) is a Hebbian learning paradigm crucial for synaptic modification.
• Burst-STDP and homeostatic renormalization are advanced learning mechanisms inspired by biological neural processes.

Purpose of the Study:

• To introduce a hierarchical network of spiking neurons for object recognition and motor control.
• To investigate the efficacy of burst-STDP and homeostatic renormalization for learning complex tasks.
• To demonstrate the compatibility of the proposed model with digital neuromorphic hardware.

Main Methods:

• Development of a hierarchical spiking neural network architecture.
• Implementation of supervised and unsupervised learning using burst-STDP with homeostatic renormalization.
• Utilizing a leaky-integrate-and-fire (LIF) neuron model with binary synapses.
• Testing the network in a simulated environment with moving objects, distractors, and noise.

Main Results:

• The spiking neural network successfully learned object recognition, motion detection, and motor output generation.
• The system demonstrated robustness in environments with distractors, concurrent object movement, and noise.
• The proposed model, using LIF neurons and binary synapses, is fully compatible with digital neuromorphic hardware.

Conclusions:

• The developed spiking neural network architecture and learning rules are promising for scalable, fully neuromorphic systems.
• The use of burst-STDP and homeostatic renormalization enables efficient learning of complex visual-motor tasks.
• The model's compatibility with digital hardware paves the way for practical implementation in neuromorphic chips.