Temporal clustering with spiking neurons and dynamic synapses: towards technological applications

Area of Science:

• Computational Neuroscience
• Machine Learning
• Signal Processing

Background:

• Temporal pattern detection is crucial for understanding complex dynamic systems.
• Existing methods may struggle with the nuances of spike timing-dependent plasticity.
• Biologically inspired neural models offer potential for advanced pattern recognition.

Purpose of the Study:

• To develop a neural network capable of detecting temporal patterns in multi-dimensional signals.
• To implement a biologically plausible model of dynamic synapses for adaptive learning.
• To demonstrate the network's function as a temporal clustering mechanism.

Main Methods:

• Utilizing a network of integrate-and-fire neurons with fully connected dynamic synapses.
• Employing a biologically plausible dynamical model for synapses based on pre- and post-synaptic spike timing.
• Configuring adaptable synapses to learn and implement specific temporal delays.

Main Results:

• Each output neuron acts as a specific temporal pattern detector.
• The network successfully performs temporal clustering, assigning outputs to input clusters.
• Demonstrated classification capabilities using Poisson processes and speech data analysis.

Conclusions:

• The proposed spiking neural network effectively detects and clusters temporal patterns.
• Dynamic synapses provide an adaptive mechanism for learning temporal sequences.
• The model shows promise for applications in signal processing and data analysis.