MLA-C01 Question 221
Select 3You are training a machine learning model using SageMaker, and you notice that the model performs very well on the training dataset but poorly on the validation dataset. What methods can you use to identify and mitigate model overfitting in this scenario?
- A
Evaluate the training and validation curves to check for a large gap in performance.
- B
Decrease the complexity of the model by reducing the number of parameters or layers.
- C
Increase the training data by using data augmentation or collecting more data.
- D
Train the model for more epochs to further improve its fit on the training data.
- E
Check for high bias through the training loss and validation loss being similar and high.
Show answer and explanation
Correct answers: A, B, C
Explanation
Overfitting occurs when a model performs well on the training data but poorly on unseen or validation data. This can be identified by analyzing a large gap between training and validation performance. To mitigate overfitting, you can reduce model complexity, increase the amount of training data, or improve generalization through regularization techniques. Training for more epochs or checking for high bias are not effective methods to address overfitting.
- A. Correct.
Evaluating the training and validation curves is a common method to identify overfitting. A large gap between training and validation performance typically indicates overfitting.
- B. Correct.
Decreasing the complexity of the model, such as reducing the number of parameters or layers, can help mitigate overfitting by reducing the model's capacity to memorize the training data.
- C. Correct.
Increasing the training data through techniques like data augmentation or collecting more data can help the model generalize better and reduce overfitting.
- D. Incorrect.
Training the model for more epochs might exacerbate overfitting because the model will likely continue to memorize the training data rather than generalize to unseen data.
- E. Incorrect.
High bias is associated with underfitting, not overfitting. Overfitting is characterized by strong performance on the training data but poor generalization to the validation data.