Fine-grained Optimization of Deep Neural Networks