Spike Timing or Rate? Neurons Learn to Make Decisions for Both Through Threshold-Driven Plasticity

Area of Science:

• Computational Neuroscience
• Machine Learning

Background:

• Spiking neurons are crucial for information processing in the central nervous system.
• Existing learning algorithms for spiking neurons often focus on either rate-based or temporal coding, but not both.
• The neuroscience community has an ongoing debate regarding the primary coding mechanism in neural information transmission.

Purpose of the Study:

• To develop a general learning algorithm for spiking neurons capable of processing both rate and temporal codes.
• To address the limitations of existing algorithms that assume a single coding strategy.
• To investigate a unified approach to neural learning that accommodates the rate vs. temporal coding debate.

Main Methods:

• Proposed several threshold-driven plasticity algorithms for training spiking neurons.
• Provided mathematical proofs for algorithm properties like robustness and convergence.
• Conducted experiments to evaluate the algorithms' performance in learning spike patterns and classification tasks.

Main Results:

• The developed algorithms are simple, effective, and efficient for training spiking neurons.
• Neurons trained with these algorithms can detect and recognize embedded features from background sensory activity.
• A single neuron can perform multicategory classifications based on output spike counts.
• The algorithms automatically extract temporal features when spike timings are relevant, without explicit instruction.

Conclusions:

• The proposed threshold-driven plasticity algorithms offer a unified approach to learning in spiking neural networks.
• These algorithms are potentially beneficial for both software and hardware implementations due to their simplicity and efficiency.
• The algorithms demonstrate the capability of spiking neurons to learn and process information encoded in both spike rates and precise timings.