Distributed Statistical Min-Max Learning in the Presence of Byzantine Agents