Spatio-Temporal Sequential Memory Model With Mini-Column Neural Network.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Neuroscience

Background:

• Human memory mechanisms remain incompletely understood, hindering the development of effective computational models.
• Existing memory models often exhibit unsatisfactory performance in complex data storage and retrieval tasks.

Purpose of the Study:

• To propose a novel brain-inspired memory model utilizing spiking neural networks (SNNs).
• To enhance memory storage capacity and retrieval accuracy through a biologically plausible architecture.

Main Methods:

• Developed a spatio-temporal sequential memory model structured by mini-columns of biological spiking neurons, inspired by the neocortex.
• Implemented Spike-Timing-Dependent Plasticity (STDP) for associative learning between memory items via excitatory neuron connections.
• Utilized inhibitory neurons to prevent erroneous predictions and improve retrieval precision.

Main Results:

• The proposed SNN model demonstrated effective storage of large datasets.
• Accurate data retrieval was achieved when sufficient contextual information was provided.
• The model's architecture, incorporating excitatory/inhibitory neurons and mini-columns, proved effective for memory formulation.

Conclusions:

• The developed brain-inspired SNN memory model offers a promising new approach to computational memory.
• The findings suggest the potential of integrating principles like STDP, spike-based encoding, and mini-column structures for advanced memory systems.