Optimizing Deep Neural Networks via Hybrid Gradient Descent

Bob Jones

Training high-parameter models suffers from slow convergence rates. We propose a hybrid optimization technique merging Adam and SGD to accelerate neural training.