Question

A machine learning engineer is training a deep neural network image classifier on SageMaker. Using a very large batch size to maximize GPU throughput, they observe that training loss converges quickly but validation accuracy is noticeably lower than a prior run that used a smaller batch size. Compute budget allows either approach. Which adjustment is MOST likely to improve generalization while keeping training stable?

Accepted Answer

Reduce the batch size (and scale the learning rate accordingly) so the model sees noisier gradient estimates that can improve generalization

Answer

Increase the batch size further to reduce gradient variance and speed convergence

Answer

Remove all regularization and dropout so the model can fit the training data more completely

Answer

Disable shuffling of the training data between epochs to make gradient updates more consistent