Watch this video on YouTube
When you observe oscillations in the loss during batch training of a neural network, how should you modify your model to ensure convergence?