Arnab Neelim Mazumder
Papers
2
Total Citations
65
H-Index
2
About
Arnab Neelim Mazumder is at the forefront of making advanced artificial intelligence practical for the smallest devices. His research centers on the optimization of deep neural networks (DNNs) and vision transformers (ViTs) for on-device inference, particularly within the resource-constrained realm of TinyML. Mazumder’s major contribution lies in bridging the gap between high-performance AI models and the severe limitations of microcontrollers and edge hardware. His highly cited survey on neural network accelerator optimization (63 citations) provides a foundational roadmap for deploying micro-AI, systematically addressing the challenges of energy, memory, and computation. Building on this, his recent work introduces ViT-Reg, a pioneering hardware-aware fine-tuning framework that enables vision transformers—typically too large for embedded systems—to operate efficiently on TinyML platforms. This methodology models performance and energy behavior to balance accuracy with resource constraints, marking a significant step toward practical, real-world edge AI. Mazumder’s research is essential reading for anyone seeking to understand how to shrink powerful AI into the palm of your hand.
Research Focus
Key Achievements
Top Papers
- 1
- 2