Precise-spike-driven synaptic plasticity: learning hetero-association of spatiotemporal spike patterns

Area of Science:

• Computational Neuroscience
• Machine Learning

Background:

• Spatiotemporal pattern processing is crucial for neural computation.
• Existing learning rules often lack biological plausibility or computational efficiency.

Purpose of the Study:

• Introduce and evaluate the Precise-Spike-Driven (PSD) Synaptic Plasticity learning rule.
• Demonstrate PSD's efficacy in processing and memorizing spatiotemporal patterns.

Main Methods:

• Analytically derived PSD rule from the Widrow-Hoff rule for supervised learning.
• Trained neurons to associate input spike patterns with desired output spike trains.
• Investigated PSD properties via simulations: performance, generality, robustness, capacity, and parameter effects.

Main Results:

• PSD rule demonstrated effective spatiotemporal pattern classification.
• Outperformed a benchmark algorithm using a relative confidence criterion.
• Achieved good recognition performance in an optical character recognition task.

Conclusions:

• PSD Synaptic Plasticity is a computationally efficient and biologically plausible learning rule.
• PSD shows strong potential for applications in neural computation and pattern recognition.
• The rule offers advantages over existing algorithms like tempotron and Chronotron.