Building Interpretable and Robust Deep Learning Models Towards Trustworthiness