Researcher at Torr Vision Group | Visiting Researcher at Sanger Institute
I am currently a Master's student at the University of Oxford (Torr Vision Group), focusing on Trustworthy Multimodal Learning and Generative Models. Previously, I graduated with a B.S. in Computer Engineering from the University of Isfahan (Top 5%).
My research aims to make AI systems more robust, consistent, and scientifically useful.
-
🔭 I’m currently working on:
- Debugging multimodal hallucination errors in VLMs via uncertainty propagation (with Prof. Yarin Gal).
- Physics-aware generative video models (with Prof. Philip Torr).
- Computational Single-Cell Genomics and 3D imputation (with Prof. Mo Lotfollahi).
-
🔬 Research Interests:
- Trustworthy & Robust ML: Uncertainty, Calibration, Privacy-Preserving ML.
- 3D Perception: Video Reasoning, 3D-consistent generation.
- Multimodal LLMs: Vision-Language benchmarks, Hallucination analysis.
- AI for Science: Spatial transcriptomics, Medical imaging.
| Paper | Venue | Description |
|---|---|---|
| Hidden in Plain Sight: Evaluating Abstract Shape Recognition in Vision-Language Models | NeurIPS 2024 | Built IllusionBench with diffusion models to audit VLM shape perception under zero/few-shot settings. |
| 3D-Guided Scalable Flow Matching for Generating Volumetric Tissue Spatial Transcriptomics | CVPR 2026 (Submitted) | HoloTea: A 3D-aware flow-matching framework for 3D-consistent volumetric spatial transcriptomics. |
| VAGUE-Gate: A Plug-and-Play Local-Privacy Shield for RAG | AACL 2025 | A client-side local differential privacy gate that rewrites context before LLM inference. |
| MEENA (PersianMMMU) | EACL 2026 (Under Review) | First large-scale Persian VLM benchmark (7.5k Qs) evaluating GPT-4, Gemini, etc. |
| From Scenes to Semantics: PersianCLEVR for Bilingual 3D Visual Reasoning | NeurIPS 2025 Workshop | Bilingual 3D visual reasoning benchmark connecting synthetic scenes to compositional queries. |
🔻 Click to see more publications & workshops
- ScenePhys - Controllable Physics Videos for World-Model Evaluation (NeurIPS 2025 Workshop - EWM)
- RAG-Driven Video QA with Adaptive Chunking (EduViQA)
- Context Awareness Gate for RAG (15th IEEE-IKT, Oral)
- Advanced Mutation Testing with Zero/Few-Shot using GPT-4 (9th IEEE-IoT, Oral)
Last updated: December 2025