Jay Gala
Research MSc @ Mila / McGill University
I am an MSc student in Computer Science at Mila and McGill University, supervised by Siva Reddy and Dzmitry Bahdanau. My research interests span data-efficient learning, training dynamics, reasoning, and generalization.
Previously, I was a Research Associate at MBZUAI working on multimodal learning with Yova Kementchedjhieva and Alham Fikri Aji, and collaborated with Zeerak Talat on hate speech detection using federated learning. Before that, I was an AI Resident at AI4Bharat (IIT Madras) where I built open-source datasets and models for Indian languages with Mitesh Khapra, Anoop Kunchukuttan, and Raj Dabre. I also interned at UC San Diego with Pengtao Xie on neural architecture search and generative models.
I hold a Bachelor’s degree in Computer Engineering from the University of Mumbai, India. Outside research, I served as a mentor at DJ Unicode, where I also led a team that built a platform for conducting C programming examinations for over 500 students (demo). I co-founded Unicode Research with Swapneel Mehta, served as a TA for the Unicode ML Summer Course 2021 supported by Google Research India, and was a founding research engineer at SimPPL, where I collaborated with The Sunday Times and Ippen Digital to build a tool for auditing online disinformation.
News and Timeline
- January 2026: Our preprint on BRIDGE: Predicting Human Task Completion Time From Model Performance is now available on arXiv.
- August 2025: Our work LLMs Can Compensate for Deficiencies in Visual Representations got accepted at EMNLP 2025.
- June 2025: Our preprint on LLMs Can Compensate for Deficiencies in Visual Representations is now available on arXiv.
- January 2025: Our work on MMTEB: Massive Multilingual Text Embedding Benchmark is accepted to ICLR 2025.
- August 2024: Gave a talk on in-context learning capabilities of LLMs for MT (slides) at the SNLP Reading Group, Microsoft Research India.
- August 2024: Our work RomanSETU received 🏆 Senior Area Chair Award at ACL 2024! Congratulations to all the authors!
- May 2024: Our works - RomanSETU, ICL study for MT and Data Pruning for MT got accepted at ACL 2024.
- May 2024: Our work on Leverage Class-Specific Accuracy to Guide Data Generation for Improving Image Classification is accepted at ICML 2024. Stay tuned for the camera-ready version!
- March 2024: Our new preprint On the low-shot transferability of [V]-Mamba is now out on arXiv.
- January 2024: Our preprint on ICL abilities in LLMs for MT is available on arXiv.
- January 2024: Excited to announce the release of Airavata, an instruction-tuned Hindi LLM. Check out the Technical Report and Code.
- November 2023: IndicTrans2 submission has been accepted at TMLR. Check out the Camera Ready Version.
- November 2023: Presenting tutorial on Developing SOTA MNMT Systems for Related Languages at AACL-IJCNLP 2023.
- May 2023: Excited to share the release of IndicTrans2, first open-source model to support all 22 Scheduled Indian languages. Check out the Preprint and Code.
- January 2023: A Federated Approach for Hate Speech Detection has been accepted to EACL 2023. Check out the Preprint and Code.