Characterizing the Optimal 0 1 Loss for Multi-class Classification with a Test-time Attacker

Oct-9-2025, 02:30:04 GMT–Neural Information Processing Systems 

Finding classifiers robust to adversarial examples is critical for their safe deployment.