Yamin Sepehri

About

Yamin Sepehri is a researcher advancing the frontiers of efficient deep learning, with a focus on bridging the gap between high-accuracy neural networks and resource-constrained edge devices. Their primary research areas include hierarchical training methodologies, early exiting mechanisms, and optimizing deep neural networks (DNNs) for distributed computing environments. Sepehri’s most notable contribution is the development of hierarchical training frameworks that leverage early exiting strategies, enabling DNNs to achieve state-of-the-art accuracy while dramatically reducing computational overhead. This work, published in 2024 and garnering 11 citations, addresses critical challenges in deploying vision models on edge devices—such as high communication costs, runtime delays, and privacy risks—by shifting training processes away from cloud servers. By designing architectures that allow intermediate layers to make early predictions, Sepehri’s research empowers efficient on-device learning without sacrificing performance. Their work has significant implications for real-time applications in IoT, autonomous systems, and mobile computing, where low-latency and data privacy are paramount. Sepehri’s innovative approach to hierarchical training positions them as a key contributor to the next generation of scalable, privacy-preserving AI systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Hierarchical Training of Deep Neural Networks Using Early Exiting

   11 citations · 2024

Key Collaborators

Contact & Links