Jehanzeb Mirza | Agentic & Multimodal AI

Hi, I am Jehanzeb Mirza. I am a Staff Research Scientist at Xero, where I build agentic AI systems for financial applications (tool use, structured reasoning, evaluation, and optimization). Previously, I was a Postdoctoral Researcher at MIT CSAIL in the Spoken Language Systems Group led by Dr. James Glass. I received my Ph.D. in Computer Science (Computer Vision) from TU Graz, Austria, where I was advised by Professor Horst Bischof, and Professor Serge Belongie served as an external referee.

My research spans multimodal foundation models (vision, language, audio) and test-time learning, with an emphasis on robust reasoning and decision-making. I am particularly interested in building reliable AI agents that can interface with tools and operate over complex structured data.

Outside of work, I enjoy traveling, running, and playing squash. I currently take squash lessons from former World No. 1 Thierry Lincou.

Selected work has been featured by MIT News and CSAIL research spotlights. I’m always happy to connect with student collaborators and researchers working on multimodal learning, LLM/VLM reasoning, and agentic systems—feel free to email me for feedback or collaboration.

Contact

jmirza [at] mit.edu
Office: 32-G442.
MIT, Cambridge, USA.

Education

Ph.D. in Computer Vision (2021 - 2024)
TU Graz, Austria.
MS in ETIT (2017 - 2020)
KIT, Germany.
BS in EE (2013 - 2017)
NUST, Pakistan

Recent News

02/26: 2 papers accepted at CVPR, 2026.

01/26: I left MIT and joined Xero as a Staff Research Scientist.

01/26: 1 paper accepted at ICLR, 2026.

10/25: Our recent ICCV work was covered by MIT News: Story.

10/25: I have been recognized as NeurIPS 2025 Exceptional Reviewer.

09/25: My recent research was covered by MIT-CSAIL: Blogpost | Video.

09/25: 1 paper accepted at NeurIPS, 2025.

08/25: 1 paper accepted at TMLR, 2025.

07/25: 1 paper accepted at COLM, 2025.

06/25: 2 paper accepted at ICCV, 2025.

04/25: Our workshops "Long Multi-Scene Video Foundations" and "MMFM" got accepted at ICCV 2025.

03/25: Talk at EI Seminar, MIT-CSAIL.

02/25: 2 paper accepted at CVPR, 2025 (workshops).

01/25: 3 papers accepted at ICLR, 2025.

12/24: Our workshop "What's Next in Multi-Modal Foundation Models" got accepted at CVPR 2025.

11/24: I joined MIT CSAIL as a Postdoctoral Researcher.

11/24: 1 paper accepted at 3DV, 2025.

09/24: 1 paper accepted at NeurIPS, 2024.

07/24: 1 paper accepted at BMVC, 2024.

07/24: 2 papers accepted at ECCV, 2024.

04/24: I successfully defended my Ph.D. thesis.

12/23: Our workshop "What's Next in Multi-Modal Foundation Models" got accepted at CVPR 2024.

10/23: Invited talk at Cohere.

10/23: Invited talk at VIS Lab, University of Amsterdam.

9/23: 1 paper accepted at NeurIPS, 2023.

9/23: Invited talk at Center for Robotics, Paris Tech.

7/23: 1 paper accepted at ICCV, 2023.

4/23: I will be attending ICVSS 2023.

3/23: 2 papers accepted at CVPR, 2023.

2/23: Reviewing for CVPR, ICCV, and TPAMI.

3/22: 2 papers accepted at CVPR, 2022.

Experience

Staff Research Scientist - Xero (USA): Agentic AI systems for finance: tool use, structured reasoning, evaluation, and optimization.01.26 - Present

Postdoctoral Researcher - MIT CSAIL (Boston, USA): Multimodal learning with speech/audio, vision, and language.11.24 - 12.25

Research Assistant - TU Graz (Graz, Austria): Self-supervised learning, test-time adaptation, and vision-language understanding.01.21 - 10.24

Research Scientist Internship - Sony AI (Tokyo, Japan): Multimodal learning with vision, language, and audio.05.24 - 08.24

Internship - Intel (Karlsruhe, Germany): Robustness of 2D/3D perception systems in adverse conditions for autonomous driving.03.19 - 08.20

Selected Publications

CVPR 2026 | Agentic VLMs

TTRV

Test-time reinforcement learning strategy for improving VLM reasoning and decision quality.

Paper Project Code

ICCV 2025 | In-Context Localization

IPLoc

Teaching VLMs to localize specific objects from in-context examples for fine-grained grounding.

Paper Project Code

TMLR 2025 | Optimization

GLOV

Using guided large language models as implicit optimizers for vision-language models.

Paper Project Code

Filter:

No publications match this combination.

TTRV: Test-Time Reinforcement Learning for Vision Language Models

Akshit Singh, Shyam Marjit, Wei Lin, Paul Gavrikov, Serena Yeung-Levy, Hilde Kuehne, Rogerio Feris, Sivan Doveh, James Glass, M. Jehanzeb Mirza

CVPR 2026

[ Paper | Project Page | Code]

Teaching VLMs to Localize Specific Objects from In-context Examples

Sivan Doveh, Nimrod Shabtay, Wei Lin, Eli Schwartz, Hilde Kuehne, Raja Giryes, Rogerio Feris, Leonid Karlinsky, James Glass, Assaf Arbelle, Shimon Ullman, M. Jehanzeb Mirza

ICCV 2025

[ Paper | Project Page | Code]

GLOV: Guided Large Language Models as Implicit Optimizers for Vision Language Models

M. Jehanzeb Mirza, Mengjie Zhao, Zhuoyuan Mao, Sivan Doveh, Wei Lin, Paul Gavrikov, Michael Dorkenwald, Shiqi Yang, Saurav Jha, Hiromi Wakaki, Yuki Mitsufuji, Horst Possegger, Rogerio Feris, Leonid Karlinsky, James Glass

TMLR 2025

[ Paper | Project Page | Code]

Are Vision Language Models Texture or Shape Biased and Can We Steer Them?

Paul Gavrikov, Jovita Lukasik, Steffen Jung, Robert Geirhos, Bianca Lamm, M. Jehanzeb Mirza, Margret Keuper, Janis Keuper

ICLR 2025

[ Paper]

Mining your Own Secrets: Diffusion Classifier Scores for Continual Personalization of Text-to-Image Diffusion Models

Saurav Jha, Shiqi Yang, Masato Ishii, Mengjie Zhao, Christian Simon, M. Jehanzeb Mirza, Dong Gong, Lina Yao, Shusuke Takahashi, Yuki Mitsufuji

ICLR 2025

[ Paper]

ConMe: Rethinking Evaluation of Compositional Reasoning for Modern VLMs

^*Irene Huang, ^*Wei Lin, ^*M. Jehanzeb Mirza, Jacob Hansen, Sivan Doveh, Victor Ion Butoi, Roei Herzig, Assaf Arbelle, Hilde Kuehne, Trevor Darrel, Chuang Gan, Aude Oliva, Rogerio Feris, Leonid Karlinsky (*Equal Contribution)

NeurIPS 2024

[ Paper | Dataset | Code]

Meta-Prompting for Automating Zero-shot Visual Recognition with LLMs

M. Jehanzeb Mirza, Leonid Karlinsky, Wei Lin, Sivan Doveh, Jakub Micorek, Mateusz Kozinski, Hilde Kuhene, Horst Possegger

ECCV 2024

[ Paper | Project Page | Code]

Towards Multimodal In-Context Learning for Vision & Language Models

Sivan Doveh, Shaked Perek, M. Jehanzeb Mirza, Amit Alfassy, Assaf Arbelle, Shimon Ullman, Leonid Karlinsky

ECCVW 2024

[ Paper]

LaFTer: Label-Free Tuning of Zero-shot Classifier using Language and Unlabeled Image Collections

M. Jehanzeb Mirza, Leonid Karlinsky, Wei Lin, Mateusz Kozinski, Horst Possegger, Rogerio Feris, Horst Bischof

NeurIPS 2023

[ Paper | Project Page | Code]

MATE: Masked Autoencoders are Online 3D Test-Time Learners

^*M. Jehanzeb Mirza, ^*Inkyu Shin, ^*Wei Lin, Andreas Schriebl, Kunyang Sun, Jaesung Choe, Mateusz Kozinski, Horst Possegger, In So Kweon, Kun-Jin Yoon, Horst Bischof (*Equal Contribution)

ICCV 2023

[ Paper | Project Page | Code]

ActMAD: Activation Matching to Align Distributions for Test-Time-Training

M. Jehanzeb Mirza, Pol Jane Soneira, Wei Lin, Mateusz Kozinski, Horst Possegger, Horst Bischof

CVPR 2023

[ Paper | Project Page | Code]

Video Test-Time Adaptation for Action Recognition

^*Wei Lin, ^*M. Jehanzeb Mirza, Mateusz Kozinski, Horst Possegger, Hilde Kuehne, Horst Bischof (*Equal Contribution)

CVPR 2023

[ Paper | Code]

The Norm Must Go On: Dynamic Unsupervised Domain Adaptation by Normalization

M. Jehanzeb Mirza, Jakub Micorek, Horst Possegger, Horst Bischof

CVPR 2022

[ Paper | Code]