Home▸AI & Machine Learning▸Neural Trojan: Backdoor Trigger Attack Lab (2D)

Neural Trojan: Backdoor Trigger Attack Lab (2D)

A top-down 2D heatmap of a classifier's decision surface: poison a training set with a hidden trigger and watch a backdoor pocket carve itself into the decision field while clean accuracy stays high.

AI & Machine Learning2DAdvanced60 FPS📱 Mobile-adapted⇄ 3D version

A top-down heatmap of a classifier's decision surface: watch a hidden backdoor carve a pocket into an otherwise honest decision boundary as poisoned, trigger-stamped samples are added. Pan and zoom the field, spin a cross-section line through it, tune poison rate, trigger strength and trigger radius, retrain, and probe a sample with and without the trigger stamp to see the attack succeed live, with Monte-Carlo clean-accuracy and attack-success-rate readouts.

⚙ Under the hood

A top-down 2D heatmap of a classifier's decision surface: poison a training set with a hidden trigger and watch a backdoor pocket carve itself into the decision field while clean accuracy stays high. Pan, zoom and slice the field with a rotating cross-section.

ai securitybackdoor attackdata poisoningneural trojanadversarial mltrust and safety

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install