Scholar
Momin Ahmad Khan
Google Scholar ID: rYs8N78AAAAJ
University of Massachusetts Amherst
Citations & Impact
Citations
45
H-index
3
i10-index
2
Publications
9
Co-authors
8
list available
Publications
Resume (English only)
Academic Achievements
- Paper 'Controlling Vision–Language–Action Policies through Sparse Latent Directions' accepted to the Mechanistic Interpretability Workshop at NeurIPS 2025; HYDRA-FL: Hybrid Knowledge Distillation for Robust and Accurate Federated Learning accepted to NeurIPS 2024; other preprints under review.
Research Experience
- Completed two internships at Nokia Bell Labs. In the second internship, focused on mechanistic interpretability for embodied AI agents—probing perception-action loops using sparse autoencoders and grounded representations. In the first internship, developed agentic LLM systems for model selection and code generation, and built a smart meeting assistant integrating LLMs, VLMs, and hardware. Currently exploring techniques like DPO and GRPO to enhance grounding and alignment in VLMs and multi-agent settings.
Education
- Completed undergraduate studies in Electrical Engineering at the School of Electrical Engineering and Computer Science at the National University of Sciences and Technology in 2021. Graduated with a gold medal for best thesis project and a silver medal for the second-highest GPA in his batch. Currently pursuing a PhD at UMass Amherst, advised by Professor Fatima Anwar.
Background
- Research interests span the security and robustness of distributed AI systems, large language models (LLMs), and vision-language models (VLMs). He has worked on designing attacks and defenses for Federated Learning (FL), identifying pitfalls in robustness evaluations, and improving prompt learning through more reliable and interpretable optimization techniques.
Miscellany
- Hobbies include learning to play the guitar, traveling alone, photography, cooking, and playing Dota 2, where he is in the top 10% of players globally.
