Publications
Blogs
Contact me
Email me
My CV
Hi, I’m
Soroush
H. Zargarbashi
PhD Researcher
CISPA Helmholtz Center in Information Security
Previous: Research Intern
Apple
Machine Translation
My research focuses on Uncertainty Quantification and Robustness of Machine Learning Models, or in general, trustworthy AI. I work extensively with conformal prediction methods to provide reliable uncertainty estimates. I have hands on experience on fine tuning large language models for speech translation.
Selected Publications
See the full list here
Under review
Optimal Conformal Prediction under Epistemic Uncertainty
Alireza Javanmardi*
Soroush H. Zargarbashi*
Aleksandar Bojchevski
Willem Waegeman
Eyke Hüllermeier
We return the smallest prediction set with conditional coverage guarantee under mild validity assumptions for second-order predictors; e.g. ensemble models, and credal set predictors.
Paper
Poster
Talk
ICML2023
Conformal Prediction Sets for Graph Neural Networks
Soroush H. Zargarbashi*
Aleksandar Bojchevski
Aleksandar Bojchevski
We leverage the homophily structure in a graph to define efficient prediction sets with valid guarantees to include the true label. Our method works on top of any confidence approach.
Paper
Poster
Talk
ICLR2024
Robust Conformal Prediction with a Single Binary Certificate
Soroush H. Zargarbashi
Aleksandar Bojchevski
We show that for symmetric randomized smoothing methods (including almost all current certificates) with only one binary certificate we can attain robust conformal prediction. Therefore we significantly reduce the number of needed Monte-Carlo samples per input.
Paper
Poster
Talk
Under review
One Sample is Enough to make Conformal Prediction Robust
Soroush H. Zargarbashi*
Sadegh Akhondzadeh
Aleksandar Bojchevski
We reduce the number of samples in randomized-smoothing based robust conformal prediction to one!
Paper
Poster
Talk
Affiliation
CISPA Helmholtz Center for Information Security
Research Assistant
2022-Ongoing
I research on Trustworthy AI, Uncertainty Quantification, Conformal Prediction, Adversarial Robustness of machine learning models.
My work is Supervised By Prof. Dr. Aleksandar Bojchevski
University of Cologne
PhD Student
2023-Ongoing
I am a teaching assistant for advanced machine learning and
My work is Supervised By Prof. Dr. Aleksandar Bojchevski
Apple
Inter, Machine Intelligence
2025: May-Oct
Researching on speech-based translation systems.
Personal Website of Soroush H. Zargarbashi | Germany | Updated Sep 2025
Soroush H. Zargarbashi
Publications
Blogs
Contact me
Hi, I’m
Soroush
H. Zargarbashi
PhD Researcher
CISPA Helmholtz Center in Information Security
Previous: Research Intern
Apple
Machine Translation
My research focuses on Uncertainty Quantification and Robustness of Machine Learning Models, or in general, trustworthy AI. I work extensively with conformal prediction methods to provide reliable uncertainty estimates. I have hands on experience on fine tuning large language models for speech translation.
Selected Publications
See the full list here
Under review
Optimal Conformal Prediction under Epistemic Uncertainty
Alireza Javanmardi*
Soroush H. Zargarbashi*
Aleksandar Bojchevski
Willem Waegeman
Eyke Hüllermeier
We return the smallest prediction set with conditional coverage guarantee under mild validity assumptions for second-order predictors; e.g. ensemble models, and credal set predictors.
Paper
Poster
Talk
ICML2023
Conformal Prediction Sets for Graph Neural Networks
Soroush H. Zargarbashi*
Simone Antonelli
Aleksandar Bojchevski
We leverage the homophily structure in a graph to define efficient prediction sets with valid guarantees to include the true label. Our method works on top of any confidence approach.
Paper
Poster
Talk
ICLR2024
Robust Conformal Prediction with a Single Binary Certificate
Soroush H. Zargarbashi
Aleksandar Bojchevski
We show that for symmetric randomized smoothing methods (including almost all current certificates) with only one binary certificate we can attain robust conformal prediction. Therefore we significantly reduce the number of needed Monte-Carlo samples per input.
Paper
Poster
Talk
Under review
One Sample is Enough to make Conformal Prediction Robust
Soroush H. Zargarbashi*
Sadegh Akhondzadeh
Aleksandar Bojchevski
We reduce the number of samples in randomized-smoothing based robust conformal prediction to one!
Paper
Poster
Talk
Affiliation
CISPA Helmholtz Center for Information Security
Research Assistant
2022-Ongoing
I research on Trustworthy AI, Uncertainty Quantification, Conformal Prediction, Adversarial Robustness of machine learning models.
My work is Supervised By Prof. Dr. Aleksandar Bojchevski
University of Cologne
PhD Student
2023-Ongoing
I am a teaching assistant for advanced machine learning and
My work is Supervised By Prof. Dr. Aleksandar Bojchevski
Apple
Inter, Machine Intelligence
2025: May-Oct
Researching on speech-based translation systems.
Personal Website of Soroush H. Zargarbashi | Germany | Updated Sep 2025