PhD student in AI reliability
I work on the reliability of deep neural networks under hardware-like faults, with a focus on software-implemented fault injection, modern vision models, and efficient evaluation methodologies. I am also interested in mechanistic interpretability: studying how models represent, process, and use information, and how those internal mechanisms affect safety.
Contact: noambires [at] gmail [dot] com
My current PhD work studies how deep neural networks behave when affected by hardware-like faults, and how to evaluate these effects with scalable software-based fault-injection workflows. I am also interested in mechanistic interpretability and model safety, especially how understanding internal model behavior can help make AI systems more reliable.
2025 - Present
Research focus: reliability of deep neural networks under hardware-like faults using software-implemented fault injection, with applications to modern vision models and efficient fault-injection methodologies. I am also building an interest in mechanistic interpretability: how models think, what internal structures drive their decisions, and how this connects to safety.
Training across artificial intelligence, computer vision, control, electrical engineering, and electronics.
2024 - 2025
2020 - 2025
First two years in EURINSA, studying in English, German, and French.
Papers section coming soon. This space will be updated with accepted publications, preprints, open-source artifacts, and reproducibility notes.
Teaching section coming soon. Future updates will include classes taught, teaching assistant roles, lab sessions, and student projects.
Outside research, I like keeping a few things in motion.