Performance Tuning in Computer Systems with Structured Bayesian Optimisation and Reinforcement Learning - Eiko Yoneki (University of Cambridge)

Abstract:

Performance tuning is a central challenge for computer systems including efficient configurations. I will introduce two recent projects: 1) Structured Bayesian Optimisation (SBO) to optimise systems in complex and high-dimensional parameter space, and 2) RLgraph, our framework for Reinforcement Learning (RL) to bring performance improvements to dynamically evolving tasks such as scheduling or resource management, where we aim at filling a gap between current research and practical deployments, and it provides a software stack for RL in systems research.

(BOAT: https://www.cl.cam.ac.uk/~ey204/pubs/2017_WWW.pdf  RLGRAPH: https://arxiv.org/abs/1810.09028, https://arxiv.org/abs/1808.07903)).

Bio:

I am an Affiliated Lecturer and Senior Researcher leading Data Centric Systems Group in the University of Cambridge Computer LaboratorySystems Research Group. I am also a Turing Fellow in the Alan Turing Institute. I received my Ph.D. degree from the University of Cambridger, 2007 (Data Centric Asynchronous Communication) and a Postgraduate Diploma in Computer Science from the University of Cambridge in 2003. Previously, I have spent several years with IBM (US, Japan, Italy and UK).