About

I am a doctoral researcher at the Indian Institute of Technology Bombay, working under the supervision of Prof. Amit Sethi in the MedaL Lab. My research lies at the intersection of federated learning, trustworthy AI, healthcare, and distributed deep learning, with a focus on building models that can learn across decentralized data silos without compromising privacy, ownership, or institutional control.

My broader research direction is motivated by a simple systems-level constraint: in high-value domains such as healthcare, scientific research, and institutional AI, data often cannot be centralized. This makes AI development not only a modeling problem, but also an infrastructure problem — requiring algorithms, runtimes, and governance-aware systems that can move computation closer to data while still achieving strong generalization and scalability.

I work on federated domain generalization, long-tailed federated learning, personalized federated models, privacy-preserving representation learning, and federated adaptation of vision-language and foundation models. Alongside my doctoral research, I serve as System Administrator for the MedaL Lab, managing NVIDIA DGX systems, Linux servers, and high-performance GPU clusters that support large-scale medical AI research.

My work has been accepted at venues and workshops associated with ICML, CVPR, AAAI, NeurIPS, ICLR, IEEE BigData, and BMVC. I have also reviewed for leading AI venues including CVPR, NeurIPS, ICLR, AAAI, ICML, and AISTATS. Beyond academia, I have received competitive research and applied-science opportunities from Mayo Clinic Rochester MN as Research Fellow, Google DeepMind, Microsoft Research, Adobe Research, Samsung R&D Institute, and Sony Research.

Current Focus

Research Alignment

My research aligns with the emerging shift from centralized AI development toward distributed, sovereign, and infrastructure-aware AI systems. As foundation models become essential infrastructure, the central question is no longer only how to build accurate models, but how to train, adapt, evaluate, and deploy them across many data owners, compute environments, and governance constraints.

I am especially interested in the intersection of:

• federated learning and distributed foundation-model training
• communication-efficient model updates and aggregation
• privacy-preserving adaptation of large models
• sovereign and institution-controlled AI deployment
• GPU/TPU-scale training infrastructure
• trustworthy AI for healthcare and scientific domains
• decentralized agentic systems and AI operating layers

This connects my PhD work in federated healthcare AI with broader infrastructure challenges in frontier AI, where data, compute, model ownership, and deployment control are increasingly distributed.

Publications

• FedNeuro: Multi-Site fMRI Analysis Using Hypernetwork Personalized and Privacy Enhanced Federated Learning

  Sunny Gupta, Shambhavi Shanker, Amit Sethi · Accepted for presentation at the AAAI 2026 Bridge Program: AI for Medicine and Healthcare (AIMedHealth), and published in the Proceedings of Machine Learning Research (PMLR), Singapore.

• FedHypeVAE: Federated Learning with Hypernetwork Generated Conditional VAEs for Differentially Private Embedding Sharing

  Sunny Gupta, Nikita Jangid, Amit Sethi · Accepted for presentation at the AAAI 2026 Bridge Program: AI for Medicine and Healthcare (AIMedHealth), and published in the Proceedings of Machine Learning Research (PMLR), Singapore.

• FedVR: Variance Regularized Hypernetwork for Federated Domain Generalization

  Sunny Gupta, Shambhavi Shanker, Amit Sethi · Oral presentation at the AAAI 2026 1st Workshop on Federated Learning for Critical Applications (FLCA @ AAAI 2026), Singapore.

• BiPrompt: Bilateral Prompt Optimization for Visual and Textual Debiasing in Vision Language Models

  Sunny Gupta, Shounak Das, Amit Sethi · Accepted for presentation at the AAAI 2026 Workshop on AI for Robust Foundation Models (AIR-FM), Singapore.

• UniVarFL: Uniformity and Variance Regularized Federated Learning for Heterogeneous Data

  Sunny Gupta, Nikita Jangid, Amit Sethi · Accepted as a Student Abstract and Poster at AAAI 2026, Singapore.

• Federated Cross-Modal Style-Aware Prompt Generation

  Sunny Gupta, Suraj Prasad, Amit Sethi · Accepted as a Student Abstract and Poster at AAAI 2026, Singapore.

• FEDTAIL: Federated Long-Tailed Domain Generalization with Sharpness-Guided Gradient Matching

  Sunny Gupta, Shounak Das, Nikita Jangid, Amit Sethi · Oral presentation at the ICML 2025 Workshop on Collaborative and Federated Agentic Workflows (CFAgentic @ ICML'25), Vancouver, Canada.

• FedAlign: Federated Domain Generalization with Cross-Client Feature Alignment

  Sunny Gupta, Vinay Sutar, Varunav Singh, Amit Sethi · Oral presentation at the 4th Workshop on Federated Learning for Computer Vision in conjunction with CVPR'25 (FedVision-2025), Nashville, USA.

• FedStein: Enhancing Multi-Domain Federated Learning Through James-Stein Estimator

  Sunny Gupta, Nikita Jangid, Amit Sethi · Accepted for presentation at the AAAI Bridge Program 2025: AI for Medicine and Healthcare (AIMedHealth), to appear in PMLR, Philadelphia, USA.

• FLeNS: Federated Learning with Enhanced Nesterov-Newton Sketch

  Sunny Gupta, Pankhi Kashyap, Pranav Jeevan, Amit Sethi · Oral presentation at IEEE BigData 2024, Washington, DC, USA.

• FedStein: Enhancing Multi-Domain Federated Learning Through James-Stein Estimator

  Sunny Gupta, Amit Sethi · Oral presentation at the International Workshop on Federated Foundation Models (FL@FM) at NeurIPS 2024, Vancouver, Canada.

• Sequential Compression Layers for Efficient Federated Learning in Foundational Models

  Navyansh Mahla, Sunny Gupta, Amit Sethi · Accepted at the Open Science for Foundation Models (SCI-FM) Workshop, ICLR 2025, Vienna, Austria.

• Taming the Tail: Leveraging Asymmetric Loss and Padé Approximation to Overcome Medical Image Long-Tailed Class Imbalance

  Pankhi Kashyap, Pavni Tandon, Sunny Gupta, Abhishek Tiwari, Ritwik Kulkarni, Kshitij Sharad Jadhav · The 35th British Machine Vision Conference (BMVC 2024), Glasgow, UK.

Experience

Program Committees & Peer Review

Education

Technical Skills

“Building AI that not only performs, but also preserves privacy, sovereignty, and control.”