Forward-Cooperation-Backward (FCB) learning in a Multi-Encoding Uni-Decoding neural network architecture

The paper introduces FCB learning, a human-inspired method for deep neural network training. It uses a Multi Encoding Uni Decoding architecture with lateral synaptic connections for cooperation. The technique demonstrates performance in dimension reduction on four datasets, preserving data granular properties in low-rank embeddings. The summary highlights loss vs. epoch plots for different models on the MNIST and FMNIST datasets, showcasing their training performance.