Dynamics of Learning in Neural Networks: A Gaussian Profile Approach

Abstract

This article explores the impact of learning on the connection strengths between nodes in a recurrent associative attractor network, utilizing a Gaussian firing profile to simulate the adaptive nature of neural connections. Initially, the network exhibits uniform connec- tion strengths, symbolizing a naive state without prior learning. Through a computational model, we demonstrate how learning processes can dynamically adjust these connec- tions, emphasizing the strengthening of associations between spatially or functionally proximate nodes. The model employs a Gaussian function to adjust connection strengths based on node proximity, effectively mimicking biological learning mechanisms. Post- learning, nodes are reordered based on their connection strengths to reveal the emergent network structure, offering insights into how neural networks can evolve to represent and recall associative memories. This study not only sheds light on the fundamental aspects of neural plasticity but also provides a computational framework for simulating learning and memory formation in artificial neural networks.