Enhancing Learning in Sparse Neural Networks: A Hebbian Learning Approach

Artificial neural networks have proven to be capable of mastering many complex tasks. However, training such networks can be extremely resource intensive. In this research, the learning rule of a neural network trained using sparse evolutionary training (SET) is extended based on Hebbian theory. A mathematical formulation of Hebbian theory, encompassing inhibitory neurons and tailored for artificial neural networks, is proposed. The resulting novel algorithm, referred to as HebbSET, exhibits enhanced performance in terms of learning speed and final accuracy on two datasets. These findings underscore the potential of incorporating neuroscientific theories to enhance the capabilities of ANNs and bridge the gap between neuroscience and AI.