You need to write a generic test to verify whether Dense Neural Network (DNN) models automatically released by your team have a sufficient number of parameters to learn the task for which they were bu

Question

Options

• ATrain the model for a few iterations, and check for NaN values.
• BTrain the model for a few iterations, and verify that the loss is constant.
• CTrain a simple linear model, and determine if the DNN model outperforms it.
• DTrain the model with no regularization, and verify that the loss function is close to zero.

• A
  5% (1)
• B
  5% (1)
• C
  14% (3)
• D
  76% (16)

Explanation

This is a capacity test - verifying the model has enough parameters (expressiveness) to fit the training data. A model with sufficient capacity, when trained without regularization, should be able to overfit the training data and drive the training loss close to zero. If the loss cannot reach near-zero even without regularization, the model is undercapacitated (too few parameters for the task). Training for a few iterations and checking for NaN values (A) detects numerical instability, not capacity. Checking for constant loss (B) detects gradient flow issues, not capacity. Comparing against a linear model (C) tests relative performance, not absolute capacity sufficiency. Only option D directly tests whether the model can memorize training data - the minimum bar for sufficient parameters.

No community discussion yet for this question.