GeRM optimizes data utilization strategies through offline reinforcement learning, outperforming other methods in multi-task learning by leveraging a Mixture-of-Experts structure.
