Hands-On Exercise: Demonstrate Overfitting and Underfitting CT-AI Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   In a hands-on exercise, which approach is most likely to produce a model that underfits?

   • AUsing a linear model to fit data that has a clear quadratic relationship.
   • BUsing a model with a very high learning rate.
   • CUsing a deep neural network with many layers on a small dataset.
   • DUsing a high-degree polynomial model with strong regularization.
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   In machine learning, what does it mean when a model is said to be overfitting?

   • AThe model performs equally well on training and test data, indicating a good balance between bias and variance.
   • BThe model is too simple to capture the underlying patterns in the training data, resulting in poor performance on both training and test data.
   • CThe model has learned the training data too well, including its noise and random fluctuations, and fails to generalize to new data.
   • DThe model has not been trained long enough to converge, so its weights are still randomly initialized.
   Reveal Solution Discussion Select an answer first
4. 13application · medium

   A team trains a neural network for image classification. The training loss decreases to 0.01, but the validation loss increases after epoch 10 and ends at 0.85. Which mitigation is most directly targeted at this problem?

   • AIncrease the number of training epochs to allow further convergence.
   • BReduce the model's capacity by removing hidden layers.
   • CAdd dropout layers and increase L2 regularization.
   • DSwitch to a smaller batch size to add noise to the gradient estimates.
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A team is building a sentiment analysis model. A simple bag-of-words logistic regression achieves 80% accuracy on both training and test sets. The business requires 90% accuracy. The team has a large unlabeled dataset of customer reviews. Which strategy is most likely to help them reach the target?

   • ACollect more labeled data and retrain the same bag-of-words model.
   • BUse the unlabeled data for semi-supervised learning or pretraining, and add n-gram features to the model.
   • CReduce the vocabulary size to the top 100 words to simplify the model.
   • DApply stronger L2 regularization to the logistic regression.
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   In a hands-on exercise, which approach is most likely to produce a model that overfits?

   • ATraining a model with strong L2 regularization.
   • BTraining a simple linear regression model on a large dataset.
   • CTraining a model with a very small number of features.
   • DTraining a high-degree polynomial regression model on a small dataset.
   Reveal Solution Discussion Select an answer first