About Me
PhD Candidate in Artificial Intelligence
Sharif University of Technology
Researcher motivated by the question “why?”, uncovering the mechanisms of complex systems through data, learning, and theory.
📄 CV (PDF)
🔗 GitHub · LinkedIn · ✉️ mirzanahal@gmail.com
Research Interests
- Shortcut learning and spurious correlations
- Reliable and trustworthy machine learning
Education
Ph.D. in Artificial Intelligence — Sharif University of Technology (2022–present)
M.Sc. in Bioinformatics — Sharif University of Technology (2019–2022)
B.Sc. in Computer Engineering — University of Tehran (2013–2018)
Selected Publications
On the Role of Implicit Regularization of Stochastic Gradient Descent in Group Robustness
International Conference on Learning Representations (ICLR 2026)
Nahal Mirzaie, A. Alipanah, A. Abbasi, A. Farzane, H. Jafarinia,
E. Sobhaei, M. Ghaznavi, A. Najafi, M. Soleymani Baghshah, M. H. Rohban
The Silent Helper: How Implicit Regularization Enhances Group Robustness
High-dimensional Learning Dynamics (ICML Workshop 2025)
Nahal Mirzaie, M. Ghaznavi, H. Oyarhoseini, A. Alipanah, E. Sobhaei,
A. Abbasi, A. Farzaneh, H. Jafarinia, P. Sharifi Sadeh,
A. Boushehrian, M. Soleymani Baghshah, M. H. Rohban
Snuffy: Efficient Whole Slide Image Classifier
European Conference on Computer Vision (ECCV 2024)
H. Jafarinia, A. Alipanah, S. Razavi, Nahal Mirzaie, M. H. Rohban
Weakly-Supervised Drug Efficiency Estimation with Confidence Score
Medical Image Computing and Computer Assisted Intervention (MICCAI 2023)
Nahal Mirzaie, M. V. Sanian, M. H. Rohban
Springer LNCS 14244, pp. 676–685
Research Experience
Rohban Lab — Sharif University of Technology (Sep 2022 – present)
Studying how implicit inductive biases affect training dynamics and gradient flow, enhancing group robustness by mitigating shortcut learning and reliance on spurious correlations.
Vikas Lab — Aalto University, Finland (Jun 2022 – Sep 2022)
Theoretical analysis of Graph Neural Networks, exploring asynchronous message-passing mechanisms beyond the 1-WL expressivity limit.
Rohban Lab — Sharif University of Technology (Sep 2019 – Feb 2022)
Weakly supervised drug efficiency estimation on the RxRx19a dataset for COVID-19 prophylactic treatment discovery, using single-cell morphological profiling.
Faghih Lab — University of Tehran (Aug 2016 – Feb 2018)
Parameterized synthesis of self-stabilizing algorithms in symmetric networks with provable closure and deadlock-freedom guarantees.
Work Experience
Data Scientist — Tapsi (Oct 2020 – Apr 2022)
- Fraud detection pipelines for commission evasion and GPS spoofing
- Passenger campaign optimization with ROI as key KPI
Bioinformatician — Pardis Gene (Dec 2018 – Aug 2020)
- Rare disease variant discovery from WGS/WES data
- Precision oncology pipelines for somatic variant detection and treatment recommendation
Awards
- RISE-MICCAI Travel Award, MICCAI 2023 (Fully funded, USD 3,000)
Academic Activities
- ICLR 2025 Workshop Organizer — Spurious Correlation and Shortcut Learning
- Co-instructor, Bioinformatics Algorithms — Sharif University of Technology