Coverage Measures for Neural Networks CT-AI Practice Questions (Page 3)

Questions 11–14

1. 11application · medium

   A team is testing a neural network for a self-driving car. They have achieved high neuron coverage on their test set, but they are concerned that the test set may not include enough edge cases like rare weather conditions. Which complementary testing technique would best address this concern?

   • AAdversarial testing with perturbed inputs
   • BIncreasing the number of neurons in the network
   • CMutation testing of the neural network model
   • DUsing a higher activation threshold for coverage
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   When applying coverage measures to assess test adequacy for a neural network, what does a low coverage value most directly indicate?

   • AThe test data contains many incorrect labels.
   • BThe network is not trained well enough to be tested.
   • CThe network will fail on all untested inputs.
   • DThe test data has not exercised a significant portion of the network's internal behavior.
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   What is a known limitation of neural network coverage measures in testing?

   • AThey can only be applied to convolutional networks.
   • BThey have a weak correlation with fault detection.
   • CThey are difficult to compute for large networks.
   • DThey require labeled test data.
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   Why do coverage measures for neural networks need to be complemented with other testing techniques?

   • ACoverage measures only assess structural aspects and may miss semantic or behavioral faults.
   • BCoverage measures are too expensive to run on large datasets.
   • CCoverage measures are not standardized across tools.
   • DCoverage measures require access to the training data.
   Reveal Solution Discussion Select an answer first