🛡️ 2D Backdoor Poisoning and Trigger Reverse-Engineering: Poison Rate, Attack Success and the Anomaly-Index Defence in a Tiny Classifier
A defensive teaching lab on backdoor poisoning of a tiny image classifier: poison rate against attack success and clean accuracy, pixel weights per class, trigger reverse-engineering with an anomaly-index verdict and unlearning, with six independent numerical checks.
⚙ Under the hood
A defensive teaching lab on backdoor poisoning of a tiny image classifier: poison rate against attack success and clean accuracy, pixel weights per class, trigger reverse-engineering with an anomaly-index verdict and unlearning, with six independent numerical checks.
Backdoor AttackData PoisoningNeural CleanseTrigger DetectionAnomaly IndexML SecurityDefensive SecurityMachine Learning
2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install