Faster CNN Training

SIGNAL ANALYSIS

Training CNNs with random mini-batches can lead to slower convergence and a weak learning signal. A*-inspired batch selection can improve this. It's unclear if this holds up outside the lab.