An ML team is dealing with a model whose accuracy on the training data keeps improving as training continues, while its accuracy on the validation data starts to worsen partway through. Which is the MOST appropriate measure to curb this overfitting?
Choosing the appropriate measure to curb overfitting for a model whose validation accuracy worsens partway through.
Weaken the regularization parameter.
Weakening regularization makes the model more complex and worsens overfitting.
What helps curb it is strengthening regularization, and this is also not the appropriate measure for this question, 'early stopping', so it is incorrect.
Increase the number of epochs even more.
Increasing the number of epochs and training longer advances overfitting to the training data even further.
When validation accuracy has started to worsen, this is counterproductive, so it is incorrect.
Apply early stopping.
Correct. Early stopping halts training at the point when validation accuracy turns toward worsening, stopping training before overfitting to the training data advances. It is an effective measure for a model whose validation accuracy worsens partway through, as in this question.
Raise the learning rate to accelerate training.
The learning rate is a setting that determines the step size of weight updates; raising it does not provide a mechanism to curb overfitting (it can actually make training unstable).
It is not the appropriate curbing measure for this question, 'early stopping', so it is incorrect.
Remember the correct answer, 'early stopping'.
・It halts training at the point when validation accuracy turns toward worsening.
・By stopping before overfitting advances, it curbs overfitting.
'Reducing the data' and 'training for longer' both worsen overfitting, and 'lowering latency' is unrelated to curbing overfitting.