Publications

Conference Papers

RODEO: Robust Outlier Detection via Exposing Adaptive Out-of-Distribution Samples

ICML, 2024

In this work, we designed a novel method to detect outliers in adversarial settings. We were able to achieve state-of-the-art results on various tasks of outlier detection by generating adaptive outliers and expose them while training the anomaly detector adversarially.

Authors: Hossein Mirzaei, Mohammad Jafari, Hamid Reza Dehbashi, Ali Ansari, Sepehr Ghobadi, Masoud Hadi, Arshia Soltani Moakhar, Mohammad Azizmalayeri, Mahdieh Soleymani Baghshah, Mohammad Hossein Rohban

Download Paper

Scanning Trojaned Models Using Out-of-Distribution Samples

NeurIPS, 2024

In this work, we designed a trojan scanning method which is robust in various aspects, including trojan attack type, label mapping, and adversarial robustness of the classifier. We further propose a version of our work which requires no access samples from the model’s training distribution.

Authors: Hossein Mirzaei, Ali Ansari*, Bahar Dibaei Nia*, Mojtaba Nafez, Moein Madadi, Sepehr Rezaee, Zeinab Sadat Taghavi, Arad Maleki, Kian Shamsaie, Mahdi Hajialilue, Jafar Habibi, Mohammad Sabokrou, Mohammad Hossein Rohban

Download Paper