Alternative to backpropagation shows promise for training deep networks

Original: RT @hardmaru: For over a decade, we’ve accepted that end-to-end backprop is the only way to train deep networks. But holding the entire net…

Source: x.com ↗