ImprovDML: Improved Trade-off in Private Byzantine-Resilient Distributed Machine Learning