Your question is Minimizing CNN Activation Memory. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do you optimize a convolutional neural network (CNN) to minimize activation memory footprint during inference for TETRAMEM?