On the Conflict of Robustness and Learning in Collaborative Machine Learning