Towards Interpretability And Robustness Of Machine Learning Models