ICASSP 2024
Killing it with Zero-Shot: Adversarially Robust Novelty Detection
Adversarial robustness for recognizing unfamiliar inputs.
Hossein Mirzaei, Mohammad Jafari, Hamid Reza Dehbashi, Zeinab Sadat Taghavi, Mohammad Sabokrou, Mohammad Hossein Rohban.
IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024.
Overview
This work combines nearest-neighbor methods with robust pretrained features to recognize unfamiliar inputs under adversarial perturbations. It evaluates how this approach improves novelty-detection reliability across benchmarks.
Citation
H. Mirzaei, M. Jafari, H. R. Dehbashi, Z. S. Taghavi, M. Sabokrou, and M. H. Rohban. “Killing it with Zero-Shot: Adversarially Robust Novelty Detection.” ICASSP, 2024.