The authors introduce a neuroscience-inspired Solo Pass Embedded Learning Algorithm (SPELA). It is a unique candidate for training and inference applications in Edge AI devices. SPELA can also cater to the need for a framework to study perceptual representation learning and formation. The algorithm has unique features such as neural priors, no weight transport, no update locking of weights, complete local Hebbian learning, single forward pass with no storage of activations, and single weight update per sample. The authors demonstrate that SPELA can perform nonlinear classification on a noisy boolean operation dataset and show high performance across MNIST, KMNIST, and Fashion MNIST datasets.
Publication date: 16 Feb 2024
Project Page: https://arxiv.org/abs/2402.09769v1
Paper: https://arxiv.org/pdf/2402.09769
