A tile-factory net has many nodes that never produced an activation…

Question

A tile-factory net has many nodes that never produced an activation above the tester’s chosen cut during the test set. Which coverage measure is that, and what does a silent region suggest?

Accepted Answer

Neuron coverage: the proportion of neurons whose output exceeded that threshold. A silent region is a hint to add inputs — not proof the net generalizes. — The share of neurons whose output exceeded a tester-chosen cut is neuron coverage. A silent region is a hint to add inputs, not proof the net generalizes. A single threshold is not already k bins, and silence does not rule out spurious correlations.

Answer

k-multisection neuron coverage, because a single threshold is already k bins.

Answer

Neuron boundary coverage, because a silent node is an output past the training min or max.

Answer

Proof of generalization, because any silent region already rules out spurious correlations.

A tile-factory net has many nodes that never produced an activation above the tester’s chosen cut during the test set. Which coverage measure is that, and what does a silent region suggest?

Short Explanation

Full Explanation