In-memory for continuous (& simpler) deep learning Part 2