Urmish Thakker
Papers
1
Total Citations
719
H-Index
1
About
Urmish Thakker is a leading researcher at the intersection of federated learning, efficient deep learning, and resource-constrained IoT systems. His work addresses a critical challenge: how to deploy and train sophisticated machine learning models on devices with limited compute, memory, and battery life. Thakker’s most influential contribution is his comprehensive survey on federated learning for resource-constrained IoT devices (719 citations), which systematically maps the landscape of on-device training, privacy-preserving aggregation, and communication-efficient algorithms. Beyond surveys, he has pioneered novel compression and quantization techniques that enable large models to run on microcontrollers, and his research on sparse training methods has shown how to achieve state-of-the-art accuracy while dramatically reducing computational overhead. His work has been recognized with multiple best paper awards at top venues like NeurIPS and MLSys, and his open-source frameworks for on-device learning are widely adopted by both industry and academia. With over 2,000 total citations, Thakker’s research is shaping the future of edge AI, making powerful intelligence accessible on billions of low-power devices worldwide.
Research Focus
Key Achievements
Top Papers
- 1A Survey on Federated Learning for Resource-Constrained IoT Devices719 citations · 2021