In-network Machine Learning
Project Overview
Challenges
Category
Disciplines
University
Year
About
The rise of machine learning (ML) in time-sensitive applications like autonomous vehicles and financial trading demands ultra-fast response times. Traditional ML frameworks often struggle with latency and performance. This research introduces in-network ML, offloading ML inference to network devices such as switches and NICs, achieving up to 800x faster response times and up to 1000x lower power consumption compared to server-based solutions. By embedding ML directly within the network, data traffic is reduced, and efficiency is enhanced for real-time AI applications
Team
Xinpeng Hong
University of Oxford
Xinpeng Hong is a final-year Ph.D. student at the University of Oxford, specializing in the intersection between computer networking and machine learning. His research focuses on leveraging programmable network hardware to accelerate computation and machine-learning-driven applications, achieving significantly lower latency, higher throughput, and greater power efficiency compared to traditional server-based solutions. This approach has wide-ranging applications in time-sensitive areas such as finance, healthcare, smart transportation, and artificial intelligence of things (AIoT).
Gallery
