Computer Engineering · Sharif University of Technology
I'm a final-year Computer Engineering undergraduate at Sharif University of Technology. My research asks how machine learning systems fail under adversarial pressure, and how to make them reliable. So far that has meant backdoor attacks and adversarial transferability in speech recognition and robustness of speech foundation models.
Proposes TransferBreaker, a unified fine-tuning framework that mitigates adversarial transferability in black-box ASR deployments.
A word-level, temporally localized data-poisoning backdoor that uses acoustic triggers for precise semantic and sentence-level manipulation of ASR output.
Improves adversarial robustness of speech foundation models through hierarchical stabilization, with Mohammad Sabokrou as corresponding author and Mohammad Hossein Rohban as co-advisor.
Studies how robustness can be transferred to specialized descendants of speech foundation models and preserved during downstream specialization. Mohammad Hossein Rohban is the corresponding author, with Mohammad Sabokrou as co-advisor.
Logical anomaly detection for industrial images, with results competitive with state-of-the-art methods.
A survey spanning face generation, expression rendering, gesture and motion synthesis, avatar creation, and texture modeling.
Advisor: Mohammad Hossein Rohban
Advisor: Mohammad Sabokrou
Advisor: Ehsaneddin Asgari
Sep 2022 – Apr 2027
B.S. in Computer Engineering
2022
Ranked 23rd among more than 150,000 participants, placing in the top 0.02%.
Volunteer teaching assistant at Sharif University of Technology for: