Hebbian Learning Algorithms for Training Convolutional Neural Networks

The concept of Hebbian learning refers to a family of learning rules, inspired by biology, according to which the weight associated with a synapse increases proportionally to the values of the pre-synaptic and post-synaptic stimuli at a given instant of time. Different variants of Hebbian rules can be found in literature. In this thesis, three main Hebbian learning approaches are explored: Winner-Takes-All competition, Self-Organizing Maps and a supervised Hebbian learning solution for training the final classification layer of a network. In literature, applications of Hebbian learning rules to train networks for image classification tasks exist, although they are currently limited to relatively shallow architectures. In this thesis, the possibility of applying Hebbian learning rules to deeper network architectures is explored and the results are compared to those achieved with Gradient Descent on the same architectures