Halting Strategy

Henry is recording a YouTube video where he wants to explain how early stopping works when training a neural network.

The fundamental insight he wants his audience to understand is how the technique decides the exact moment it needs to halt the training process.

Assuming that Henry will use the validation loss as the metric to watch, which of the following is the correct way to configure early stopping?

The training process should stop once the validation loss has increased for several consecutive iterations.

The training process should stop once the validation loss has decreased for several successive iterations

The training process should stop once the validation loss hits its maximum value.

The training process should stop once the validation loss decreases from its immediate previous value.