A manufacturing company asks its machine learning specialist to develop a model that classifies defective parts into one of eight defect types. The company has provided roughly 100,000 images per defe

Question

Options

• AA longer training time
• BMaking the network larger
• CUsing a different optimizer
• DUsing some form of regularization

• A
  3% (1)
• B
  6% (2)
• C
  9% (3)
• D
  82% (27)

Explanation

This is a classic overfitting scenario: the model performs well on training data (90%) but generalizes poorly to unseen validation data (80%). The 10-point gap between training and validation accuracy is the key signal. Regularization techniques - such as L1/L2 weight regularization, dropout layers, data augmentation, or early stopping - directly combat overfitting by preventing the model from memorizing the training set. Option A (longer training) would make overfitting worse, not better. Option B (larger network) increases model capacity, which typically increases overfitting. Option C (different optimizer) may improve convergence speed but does not address the fundamental overfitting problem. The fact that human-level performance is ~90% (matching training accuracy) confirms the model capacity is sufficient - the problem is generalization, not underfitting. Regularization is the correct tool to close the train/validation gap.

No community discussion yet for this question.