Rohit Lal
AI for Science · Foundation Models · Computer Vision · Computer Scientist II at NASA - ODSI (Office of Data Science) / UAH
I am a Computer Scientist at NASA’s Office of Data Science and Informatics (ODSI), where I build and scale foundation models for science. I work with satellite imagery, instrument observations, and decades of atmospheric reanalysis, turning scientific data into open models that researchers can adapt to their own problems. My work includes Prithvi WxC, a foundation model for weather and climate, and Surya, a foundation model for heliophysics. I work across the full pipeline: curating datasets, designing architectures suited to the data, and scaling pretraining across hundreds of GPUs. My background is in computer vision. I completed my MS at UC Riverside, working with Prof. Amit K. Roy-Chowdhury in the Visual Computing Group on 3D human pose estimation under occlusion for the IARPA BRIAR project. Before that, I worked with Prof. Anirban Chakraborty at IISc Bangalore’s Visual Computing Lab on domain adaptation and adversarial robustness. Away from the GPUs, I am usually traveling, chasing good light with my camera, and taking table tennis a little too seriously.
Profiles
Publications
- Surya: A Foundation Model for Heliophysics. Sujit Roy, Johannes Schmude, Rohit Lal, Vishal Gaur, Marcus Freitag, Julian Kuehnert, … , Juan Bernabe-Moreno, Rahul Ramachandran .. arXiv preprint (NASA × IBM open release), 2025.
- Prithvi WxC: Foundation Model for Weather and Climate. Johannes Schmude, Sujit Roy, Will Trojak, Johannes Jakubik, …, Rohit Lal, … , Manil Maskey, Tsengdar J Lee, Rahul Ramachandran. arXiv preprint (NASA × IBM open release), 2024.
- Layer-wise Alignment: Examining Safety Alignment Across Image Encoder Layers in Vision Language Models. Saketh Bachu, Erfan Shayegani, Rohit Lal, Trishna Chakraborty, Arindam Dutta, … Nael Abu-Ghazaleh, Amit K. Roy-Chowdhury. ICML 2025 (Spotlight - top 2.6%).
- STRIDE: Single-Video based Temporally Continuous Occlusion-Robust 3D Pose Estimation. Rohit Lal, Saketh Bachu, Yash Garg, Arindam Dutta, Calvin-Khang Ta, D. S. Raychaudhuri, Hannah D. Cruz, M. Salman Asif, Amit K. Roy-Chowdhury. WACV 2025 (Oral).
- VOccl3D: A Video Benchmark Dataset for 3D Human Pose and Shape Estimation under Real Occlusions. Yash Garg, Saketh Bachu, Arindam Dutta, Rohit Lal, Sarosij Bose, Calvin-Khang Ta, M. Salman Asif, Amit K. Roy-Chowdhury. ICCV 2025.
Experience
Computer Scientist II, NASA - ODSI (Office of Data Science) / UAH (May 2024 - Present)
Building and scaling AI Foundation Models for Science.
- Weather Foundation Model: Contributed to training Prithvi WxC, a 2.3B-parameter open-source weather and climate model on NASA MERRA-2 data; designed fine-tuning for autoregressive rollout forecasting, downscaling, and gravity-wave flux parameterization. [NASA report]
- Helio Foundation Model: Built Surya, the first heliospheric foundation model on 300TB of Solar Dynamics Observatory data — explored Swin, Hiera, MaxViT, and multi-modal architectures. Close collaboration with IBM, Microsoft, and NVIDIA.
- Scaling: Trained these foundation models across 200+ A100 GPU clusters (Jülich supercomputing, DGX) with SLURM, PyTorch FSDP, Lightning, and Hugging Face Accelerate.
Graduate Student Researcher, Visual Computing Group, University of California, Riverside (Jan 2023 - Dec 2023)
MS by Research with Prof. Amit K. Roy-Chowdhury.
- Pose Estimation: State-of-the-art results on occluded human pose estimation for the STRIDE project, supporting downstream motion correction, prediction, and infilling.
- Biometric Recognition: Built unsupervised pipelines for 3D/2D human pose, body-part segmentation, and silhouette extraction under occlusion for the IARPA BRIAR program — joint work with USC, Northwestern, and EPFL.
- Trustworthy AI: TA for EE260B covering reinforcement learning, out-of-distribution generalization, explainability, uncertainty estimation, safety, and robustness.
Research Assistant, Visual Computing Lab, Indian Institute of Science (IISc), Bangalore (Jul 2021 - Jul 2022)
At the Visual Computing Lab with Dr. Anirban Chakraborty.
- Source-free Domain Adaptation: Proposed CoNMix, achieving state-of-the-art source-free single- and multi-target DA across standard benchmarks.
- Domain Gap Reduction: Introduced an unsupervised domain-gap reduction method that mitigates pseudo-label noise. (IJCV)
- Adversarial Vulnerability: Developed a holistic adversarial-vulnerability score enabling data-efficient knowledge distillation. (CVPR-W)
Deep Learning Research Intern (Remote), NUS - National University of Singapore (Apr 2020 - Nov 2020)
Completed my intern under Prof. Hongliang Ren at Medical Mechatronics Lab. My task was to do tracking of gaits generated by Origami robots. This task has been done using traditional CV techniques and the next step is to incorporate various deep learning techniques for 6D pose estimation.
Projects
- Person Follower Mobile Robot (code) (paper)
- Indian Number Plate Recognition (code)
- Solving Taxi v3 of OpenAi Gym (code)
- Self balancing Camera Platform (code)
- Word Embedding Generation using NLP (code) (paper)
- Real Time Handwritten Digit recognition (code)
- Harmonic Motion Analyser using MATLAB (code)
- Hand Gesture controlled Robot (code)
- Classifier without High Level APIs (code)
Recognition
- NASA Group Achievement Award (January 2026): Honored with NASA's Group Achievement Award for pioneering work on AI Foundation Models at NASA Marshall Space Flight Center.
- Research Impact Award (April 2026): Surya Foundation Model Team recognized with Research Impact Award at the UAH Sapphire Excellence Awards Gala.
- Outstanding Reviewer - ICCV'25 (October 2025): Recognized as Outstanding Reviewer at the International Conference on Computer Vision for exceptional peer review contributions.
- ICML Spotlight Paper (July 2025): Layer-wise Alignment paper accepted as SPOTLIGHT presentation at ICML 2025, ranking in the top 2.6% of all submissions.
- Smart India Hackathon Winner (August 2020): First position at Smart India Hackathon 2020 conducted by MHRD, winning cash prize of ₹1,00,000.
News
- Apr'26: Our Surya Foundation Model Team has received the Research Impact Award
- Mar'26: Delivered a lightning talk on an AI foundation model and was on the Deep Partnership (with NVIDIA) panel at the NAIRR Pilot’26 Annual Meeting.
- Jan'26: Our AI Foundation Model team was honored with NASA’s Group Achievement Award at an award ceremony held at NASA Marshall Space Flight Center. LinkedIn Post
- Oct'25: Recognized as an Outstanding Reviewer at ICCV’25
- Aug'25: Released Surya - Foundation model for HelioPhysics [NASA] [IBM] [ArXiv].
- Jul'25: One paper Layer-wise Alignment has been accepted as a SPOTLIGHT at ICML 2025 (Top 2.6%).
- Jun'25: One paper VOccl3D: A Video Benchmark Dataset for 3D Human Pose and Shape Estimation under real Occlusions has been accepted at ICCV 2025.
- Apr'25: Serving as a reviewer for CVPR 2025, ICLR 2025 and ICCV 2025.
- Feb'25: Delivered an oral presentation of my Master’s thesis STRIDE: Single-video based Temporally Continuous Occlusion-Robust 3D Pose Estimations at WACV 2025.
- Sep'24: Our team released Prithvi WxC, a new general-purpose AI model for weather and climate. [NASA Article] [IBM Blog] [HuggingFace CEO interview]
A complete plain-text version of this profile for language models is available at /llms.txt.