MLA-C01 Question 160
Select 3You are training a deep learning model on a large dataset using Amazon SageMaker. The model consistently performs poorly on the validation set despite lowering the learning rate and increasing the number of epochs. Upon further inspection, you notice that the model is overfitting the training data. What changes should you consider to address the overfitting issue?
- A
Increase the batch size to reduce the variance in gradient updates.
- B
Reduce the number of epochs to limit overfitting to the training data.
- C
Add dropout layers to the model to prevent co-adaptation of neurons.
- D
Decrease the batch size to improve generalization.
- E
Use data augmentation to artificially increase the size of the training dataset.
Show answer and explanation
Correct answers: B, C, E
Explanation
Overfitting occurs when a model performs well on the training data but poorly on unseen data. To address overfitting, you can reduce the number of epochs to stop training early, add regularization techniques such as dropout layers, or use data augmentation to increase the effective size and variability of the training dataset. These strategies help the model generalize better to unseen data.
- A. Incorrect.
Increasing the batch size can stabilize gradient updates, but it does not directly address overfitting. This option is not correct.
- B. Correct.
Reducing the number of epochs can halt training before the model overfits the training data, making this a valid choice.
- C. Correct.
Adding dropout layers introduces regularization, which helps prevent overfitting by reducing the dependency between neurons. This option is correct.
- D. Incorrect.
Decreasing the batch size might increase noise in gradient updates but does not directly improve generalization related to overfitting. This option is not correct.
- E. Correct.
Using data augmentation increases the diversity of the training data, which can improve generalization and reduce overfitting. This option is correct.