How would you explain the trade-off between a high learning rate and a low learning rate when training a model using SGD?
