A simpler and faster biological learning framework for increasing robustness

Abstract

Hebbian Learning has been proposed for many years. The advantage of Hebbian learning is more plausible compared with Backpropagation in the aspect of biological learning. In this work, a new Hebbian Learning Framework (HLF) is designed. From the experiment results, the proposed HLF is much simpler and faster than the state-of-the-art Hebbian learning method. In this case, it can promote the usage scenarios of Hebbian Learning. Robustness of learning algorithms remains an important problem to be solved from both the perspective of adversarial attacks and improving generalization. Another work of this dissertation is that we investigate the robustness of the proposed HLF in depth. We find that Hebbian learning based algorithms outperform conventional learning algorithms like CNNs by a huge margin of upto 18% on the CIFAR-10 dataset under the addition of adversarial noise. We highlight that an important reason for this is the underlying representations that are being learned by the learning algorithms.