Choose batch size understanding its effect on gradient noise, memory, throughput, and generalisation, and adjust the learning rate with it. Use when scaling training or running out of memory.
