This approach sometimes works, but not always. For a simple example, if you are solving control problems/ or Reinforcement Learning, you need to keep your batch

This approach sometimes works, but not always. For a simple example, if you are solving control problems/ or Reinforcement Learning, you need to keep your batch sizes small to keep latency low.