List your company

Backdoor Attack Detection on Atari Pong

87.9AUROCSTRIP
45.88456.79267.778.608Jun 11, 2026

Evaluation results

MethodLinks
STRIPStrategy=Original2026.0687.9——
PolicyGuardStrategy=Hard-coded2026.0686.1——
PolicyGuardStrategy=Original2026.0685.7——
PolicyGuardAttack Type=Q-Inception, Training steps=~40M2026.0684.310.175.3
SCALE-UPStrategy=Original2026.0683.2——
SHINEStrategy=Hard-coded2026.0681.7——
SHINEStrategy=Original2026.0681.1——
NCStrategy=Original2026.0669.4——
B3D (Black-box NC)Strategy=Hard-coded2026.0650——
SCALE-UPStrategy=Hard-coded2026.0649.5——
STRIPStrategy=Hard-coded2026.0647.5——