Machine Learning

Data poisoning changes the rules a model learns. A backdoor hides that poison behind a specific trigger, so the averages never move.