Even in reinforcement learning you can train multiple model with different data-sets concurrently and combine them for the next iteration.