A company uses a neural network and retrains it when performance drops. The current training job uses SageMaker distributed data parallelism (DDP) and takes several hours. What change will reduce total training time?

Increasing the number of training instances for SageMaker distributed data parallelism (DDP) will reduce total training time. DDP works by splitting the data across multiple instances, allowing each instance to process a subset of the data in parallel. More instances mean more parallel processing, leading to faster completion of each epoch and thus a shorter overall training time. Increasing epochs, neurons, or layers would generally increase training time because the model has more computations to perform or more iterations to complete.