Effectiveness of Biologically Inspired Neural Network Models in Learning and Patterns Memorization.

Area of Science:

• Computational Neuroscience
• Machine Learning

Background:

• The Bienenstock-Cooper-Munro (BCM) model is a key framework for understanding synaptic plasticity in neurons.
• Its application has been largely limited to neuroscience simulations, with fewer uses in data science.

Purpose of the Study:

• To implement an enhanced Bienenstock-Cooper-Munro (BCM) model by integrating classical neuroscience principles with modern deep learning.
• To improve the convergence efficiency and applicability of the BCM model in data science tasks.

Main Methods:

• Combined the original BCM plasticity rule with deep learning optimization tools.
• Performed numerical simulations on standard benchmark datasets to evaluate model performance.
• Introduced neuronal competition within the BCM network to modulate memorization and selectivity.

Main Results:

• Demonstrated the BCM model's efficiency in learning, pattern memorization, and feature extraction.
• Confirmed that BCM neuron selectivity indicates internal feature extraction, beneficial for clustering and classification.
• Showcased how neuronal competition influences memorization capacity and model selectivity.

Conclusions:

• The improved BCM model offers a viable alternative to standard machine learning methods for feature selection and classification.
• The integration of deep learning enhances the BCM model's practical utility beyond traditional neuroscience simulations.