NCA-GENM Question 17
Select 3A team is building a multimodal generative AI model that requires both image and text data as input. During training, they notice that the model performs poorly on unseen data despite achieving high accuracy on the training set. Which of the following actions could help address this issue?
- A
Perform cross-validation to evaluate model performance on different subsets of the data.
- B
Add more features by creating new engineered features from the existing data.
- C
Reduce the complexity of the model to avoid overfitting the training data.
- D
Use a larger batch size during training to improve generalization.
- E
Incorporate early stopping to prevent the model from excessively fitting the training data.
Show answer and explanation
Correct answers: A, C, E
Explanation
The poor performance on unseen data indicates overfitting. Cross-validation (Option 1) helps identify overfitting by testing the model on multiple validation sets, reducing the risk of relying solely on training accuracy. Reducing model complexity (Option 3) makes the model less prone to memorizing the training data. Early stopping (Option 5) halts training when further iterations no longer improve validation performance, preventing excessive fitting to the training data. Adding more features (Option 2) and increasing batch size (Option 4) are less relevant to addressing overfitting in this scenario.
- A. Correct.
Cross-validation helps assess how well the model generalizes to unseen data by splitting the dataset into multiple training and validation subsets. It can reveal overfitting issues and suggest if the model is not generalizing well.
- B. Incorrect.
While adding more features might sometimes improve the model, it can also introduce noise or redundancy if done without careful consideration, worsening overfitting issues.
- C. Correct.
Reducing the complexity of the model (e.g., by limiting the number of layers or neurons) is an effective way to address overfitting, as simpler models are less likely to memorize the training data.
- D. Incorrect.
Using a larger batch size does not directly address overfitting. It primarily affects the stability and speed of training but does not necessarily improve model generalization.
- E. Correct.
Early stopping is a common and effective technique to mitigate overfitting by halting training once the model's performance on validation data stops improving.