Proximal SCOPE for Distributed Sparse Learning: Better Data Partition Implies Faster Convergence Rate