Question

A data science team has finished developing a fraud-detection neural network. During development, they repeatedly ran the dataset through the model many times, adjusting internal weights to minimize prediction error over several days on a multi-GPU cluster. Now they want to deploy the finalized model so a banking application can score incoming transactions in real time. Which statement best characterizes the difference between the two phases they are moving between?

Accepted Answer

The development phase is training, which iteratively updates model weights and is compute-intensive; the deployment phase is inference, which uses the fixed trained model to make predictions on new data.

Answer

Both phases are inference; the difference is only that deployment uses a larger batch size to improve accuracy.

Answer

The development phase is inference because it runs the dataset through the model, while deployment is training because it processes live production data.

Answer

The development phase requires no GPUs, while deployment always requires more GPUs than training to meet latency targets.