What is the name of the training method that selects the optimal iteration in model optimization?

The training method that selects the optimal iteration in model optimization is commonly referred to as stopped training. This approach involves monitoring the model's performance on a validation dataset during the training process and determining when to halt training to prevent overfitting. By observing how the model's validation performance changes over time, one can identify the point at which further iterations no longer provide significant improvements, thereby stopping the training process at the optimal iteration.

Stopped training ensures that the model retains its ability to generalize well to unseen data, rather than being overly fitted to the training dataset. This technique is crucial in maintaining the balance between bias and variance, which ultimately enhances the model’s predictive performance.

The other options do not capture this concept accurately. Incremental training refers to updating the model in stages, typically with new data, rather than selecting a specific point in the training. Convergence training is not a widely recognized term in the context of training methodologies; it generally implies reaching a satisfactory level of performance rather than optimizing iterations. Path training suggests following a specific route through the parameter space, which does not directly correlate with the decision of when to stop training based on performance metrics.