Do switches dream of machine learning?

The paper "Do switches dream of machine learning?" appeared at the HotNets 2019 conference last week. The research paper explores the potential use of commodity programmable switches for in-network classification.

The paper is based on an MPhil project from 2018/19, by Zhaoqi Xiong. Dr Noa Zilberman co-authored the paper, and also gave a talk at the Department of Computer Science and Technology on 7th November as part of the Systems Research Group seminar. Dr Zilberman presented the paper in Princeton, New Jersey, during the HotNets conference on 14-15th November.

As machine learning comes to dominate increasing aspects of our digital lives, there are increasing demands on the systems supporting machine learning. These challenges have driven innovation within the systems community for hardware design, including CPU optimisation, GPU and FPGA solutions. The paper concentrates on "classification for ML inference that is not neural network based".