US11601468B2 - Detection of an Adversarial Backdoor Attack on a Trained Model at Inference Time