Free Amazon MLS-C01 Actual Exam Questions - Question 3 Discussion
An insurance company is creating an application to automate car insurance claims. A machine
learning (ML) specialist used an Amazon SageMaker Object Detection - TensorFlow built-in algorithm
to train a model to detect scratches and dents in images of cars. After the model was trained, the ML
specialist noticed that the model performed better on the training dataset than on the testing
dataset.
Which approach should the ML specialist use to improve the performance of the model on the
testing data?
A imo, increasing momentum can help the model escape local minima and improve generalization a bit, though it’s less direct than regularization tweaks. Since the issue is better training performance but worse testing, overfitting is likely, so regularization methods are usually best. But if you don’t want to mess with regularization right away, tweaking momentum might smooth out training and improve test results slightly without risking underfitting. It’s not as strong a fix as D, but still worth considering alongside other hyperparameters.
Good point on overfitting—D makes sense since L2 adds regularization.
Makes sense to me that dialing down the learning rate (C) could help generalize better on test data.
I'd rule out B since reducing dropout usually lowers regularization, making overfitting worse. So options that increase regularization like D or tweaking learning rates seem more logical here.
Guessing D, adding L2 regularization usually helps with overfitting issues.
Totally agree, increasing L2 (D) is classic for tackling overfitting here.
B/D? Since the model is doing better on training than testing, it’s likely overfitting. Increasing L2 (D) makes sense to regularize, but reducing dropout rate (B) actually reduces regularization, which might make overfitting worse. So B probably isn’t right. The best move to tackle overfitting seems to be increasing regularization, so D is a stronger choice here.
Maybe D makes sense since adding L2 regularization helps reduce overfitting by penalizing complexity, which matches the issue of better train vs test performance here.
D imo, increasing L2 regularization helps prevent overfitting by adding a penalty on large weights, which should improve test performance compared to training.
Option D