Interpretability
Tracing and explaining computations inside modern models.
Mechanistic Interpretability
We develop intervention-based tools to trace how semantic functions form and transfer across layers in language and vision-language models.
Vision Transformer Attribution
We introduce more stable and faithful attribution techniques that address known failure modes in transformer explanations.
Publications
-
ACL Findings 2026
Mechanistic Interpretability of Large-Scale Counting in LLMs through a System-2 Strategy
Hosein Hasani, Mohammadali Banayeeanzade, Ali Nafisi, Sadegh Mohammadian, Fatemeh Askari, Mobin Bagherian, Amirmohammad Izadi, Mahdieh Soleymani Baghshah
Reveals how a System-2 strategy decomposes large counting problems, transfers partial counts through dedicated attention heads, and aggregates them reliably.
Findings of the Association for Computational Linguistics: ACL, 2026
-
arXiv 2026
Causal Attribution via Activation Patching
Amirmohammad Izadi, Mohammadali Banayeeanzade, Akbar Mirrokni, Hosein Hasani, Mobin Bagherian, Faridoun Mehri, Mahdieh Soleymani Baghshah
Develops activation-patching methods for tracing causal contributions inside neural networks rather than relying on correlational explanations.
arXiv preprint, 2026
-
CVPR 2025
LibraGrad: Balancing Gradient Flow for Universally Better Vision Transformer Attributions
Faridoun Mehri, Mahdieh Soleymani Baghshah, Mohammad Taher Pilehvar
Balances gradient flow to produce more stable, faithful, and broadly effective attribution maps for vision transformer architectures.
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025
-
arXiv 2025
Understanding Counting Mechanisms in Large Language and Vision-Language Models
Hosein Hasani, Amirmohammad Izadi, Fatemeh Askari, Mobin Bagherian, Sadegh Mohammadian, Mahdieh Soleymani Baghshah
Compares internal counting strategies across language and vision-language models to reveal shared mechanisms and modality-specific failures.
arXiv preprint, 2025