What technique involves halting model training before the training loss has completely decreased?

Early stopping is a technique used in training machine learning models that involves monitoring the model's performance on a validation set during training and halting the training process before the training loss has fully converged. The primary goal of early stopping is to prevent overfitting, which occurs when a model learns the training data too well, including its noise and outliers, thereby negatively impacting its ability to generalize to new, unseen data.

By stopping the training process at an optimal point—often when the validation loss starts to increase while the training loss continues to decrease—early stopping ensures that the model retains a balance between bias and variance. This approach not only saves computational resources but also often leads to better model performance on real-world data by maintaining the model's ability to generalize.

The other techniques mentioned, such as model pruning, dropout, and gradient clipping, serve different purposes within the training process and are not specifically concerned with the timing of halting training to manage overfitting or model performance related to the validation dataset.