ConceptTheory & Algos
Mitigation as canonization
Mitigation produces a backdoor-free model whose output distribution is statistically close to that of a canonical model sampled independently of the backdoor from ground truth.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Backdoored models1Total variation distance1
Lecturers
1Connections
2The lectures that define it or assume it, and the concepts next to it.