In our framework, an RL rule is represented by a meta-network that determines the targets towards which the agent should move its predictions and policy (Fig. 1c). This allows the system to discover ...