Text this: Hardware implementation of backpropagation using progressive gradient descent for in situ training of multilayer neural networks.