Meta‐learning for encoder selection