Unit 7 - Model Evaluation

What is the purpose of the validation set in evaluating a CNN model?

• To test the model's performance on data it has never seen during training
• To calculate the final accuracy of the model
• To fine-tune the model's hyperparameters
• To calculate the training loss

Which evaluation metric is commonly used for binary classification problems when assessing a CNN's performance?

• Mean Absolute Error (MAE)
• F1 Score
• Mean Squared Error (MSE)
• R-squared (R2)

In CNN image classification, what does the term "confusion matrix" represent?

• A matrix that visually confuses the model
• A matrix that measures the similarity between input and output images
• A matrix that summarizes the true positive, true negative, false positive, and false negative predictions
• A matrix that represents the confusion between training and validation datasets

What is the purpose of k-fold cross-validation when evaluating a CNN model's performance?

• To fine-tune the model's architecture
• To validate the model on k different datasets
• To obtain multiple performance estimates by dividing the data into k subsets
• To compare different types of neural networks

When assessing the generalization of a CNN model, what is "overfitting"?

• The model's inability to learn from the training data
• The model's underfitting of the validation data
• The model's good generalization to new data
• The model's fitting the training data too closely, leading to poor performance on new data

Which metric is commonly used to assess the overall performance of a multiclass CNN classification model?

• Precision
• Recall
• F1 Score
• Mean Absolute Error (MAE)

When evaluating the performance of a CNN model for object detection, which metric measures the accuracy of localizing objects within an image?

• Classification Accuracy
• Intersection over Union (IoU)
• Precision
• Recall

Which of the following is NOT a common technique for combating class imbalance when evaluating a CNN model?

• Oversampling the minority class
• Undersampling the majority class
• Adding noise to the training data
• Using different evaluation metrics

What is the purpose of a ROC curve (Receiver Operating Characteristic) in CNN evaluation?

• To evaluate a model's performance on a test dataset
• To visualize the trade-off between true positive rate and false positive rate
• To measure the model's loss function
• To calculate the model's accuracy directly

In CNN model evaluation, what does the term "precision" measure?

• The ability to correctly classify positive cases
• The ability to correctly classify negative cases
• The overall accuracy of the model
• The ability to correctly classify both positive and negative cases