How do you optimize a convolutional neural network (CNN) to minimize activation memory footprint during inference for TETRAMEM?
