NoProp: Training Neural Networks without Back-propagation or Forward-propagation Paper • 2503.24322 • Published Mar 31 • 2