Low-memory convolutional neural networks through incremental depth-first processing

Jonathan Binas, Yoshua Bengio

We introduce an incremental processing scheme for convolutional neural network (CNN) inference, targeted at embedded applications with limited memory budgets. Instead of processing…