Can Local Learning Match Self-Supervised Backpropagation?
Researchers have developed new local self-supervised learning (SSL) algorithms that can approximate the performance of global backpropagation-based SSL in deep neural networks. These novel algorithms, particularly variants using the CLAPP loss function, show improved similarity to global backpropagation updates and achieve competitive results on image datasets. This work bridges the gap between theoretical local SSL and practical global SSL methods. AI
IMPACT Advances local learning algorithms, potentially enabling more efficient training of deep neural networks without full backpropagation.