A Control Theoretic Framework for Adaptive Gradient Optimizers in Machine Learning