During test, some nodes fire higher than any training maximum or lo…

Question

During test, some nodes fire higher than any training maximum or lower than any training minimum. Which coverage measure is that, and what caution should the tester remember?

Accepted Answer

Neuron boundary coverage (NBC): the proportion of neurons that went past the training min or max. Structural coverage is a complement, not a sufficiency proof of generalization. — Nodes that fire higher than any training maximum or lower than any training minimum are neuron boundary coverage: the share that went past the training min or max. Structural coverage is a complement, not a sufficiency proof of generalization.

Answer

Neuron coverage only, because a past-range firing is the same as one tester-chosen threshold.

Answer

k-multisection neuron coverage, because a training min or max is already k bins.

Answer

Proof the net generalizes, because NBC alone rules out spurious correlations.

During test, some nodes fire higher than any training maximum or lower than any training minimum. Which coverage measure is that, and what caution should the tester remember?

Short Explanation

Full Explanation