Towards Transparent and Automatic Deep Learning: Advances in Interpretability and Learning-Rate-Free Optimization