Activation function
A non-linear function applied to a layer's output, such as ReLU or a sigmoid, without which stacking layers would just compose into one linear transformation regardless of depth. It is what lets depth actually add representational power.
Why exams ask this
Tested as a "why does this network fail to learn" debugging question. A network with no non-linear activation collapses into a single linear transformation no matter how many layers it has, which is the specific failure the exam is checking you can diagnose.
Relevant to
Related concepts
Resources
No resources linked to this concept yet.