Palestra: Bayesian Optimization of Gaussian Processes with Applications to Performance Tuning

Track: Data Science Aplicada

Sala: Pequim

Horário: 2:05pm - 2:50pm

Dia da semana: Quarta-feira

Nível: Intermediário

Persona: Arquiteto(a), Cientista de Dados, Desenvolvedor(a) Programador(a), Desenvolvedor(a) Sênior, DevOps, Gerente de Operações, Líder Técnico(a)

Apresentação em Inglês

Compartilhe essa página no:

Pontos Principais

  • Mathematical basics of Gaussian Processes (GP);
  • Application of Bayesian Optimization (BO) to optimize high-dimensional systems;
  • Large-scale performance tuning can be automated by leveraging BO of GP, but we have to be careful how we do it.

Resumo

In 1951, Daniel Krige, a mining engineer in South Africa, invented a statistical technique for finding minerals with the fewest holes drilled for prospecting. Since drilling and sample analysis is expensive, he wanted to find the place where most minerals existed with minimal drilling and analysis effort. Over the years this method has been refined, most notably by Matheron, Mockus and Jones, with more modern statistical techniques. Today, Bayesian Optimization of Gaussian Processes is used in many engineering disciplines to efficiently explore vast design spaces. It is also used for hyperparameter optimization of neural networks.

After a brief description of the technique, we show how it can be applied to a modern microservices architecture to optimize its performance. We motivate why this is an important but difficult problem, and why Bayesian Optimization is well-suited to solving this problem. We describe our implementation of a service, called Autotune, for this purpose, what issues we had to address when applying this technique, and how it will be used at Twitter to continuously optimize performance in the data-center. Some recent wins from Autotune will be highlighted.

No prior background in Bayesian Optimization, Gaussian Processes, Statistics, or Performance Tuning is needed to attend this presentation.

Palestrante: Ramki Ramakrishna

Staff Engineer at Twitter

Ramki Ramakrishna is a staff software engineer in the Platform Division of Twitter in San Francisco. He is a member of the JVM Platform team and of the Twitter Architecture Group. Ramki has worked with several generations of the JVM, at Sun and Oracle, before Twitter. He has been a committer and reviewer for the HotSpot group in OpenJDK. His principal contributions have been in the areas of performance analysis, tuning and adaptive optimization, parallel and concurrent garbage collection, and the synchronization infrastructure within the JVM. Before joining industry, Ramki worked at SUNY Stony Brook, the Tata Institute of Fundamental Research in India, and Aalborg University in Denmark, dividing time between teaching and research into the formal verification of concurrent systems, using process algebras, temporal logics and automatic theorem-proving. Ramki holds a Ph.D. in Electrical and Computer Engineering from the University of California at Santa Barbara, and a B.Tech. in Electrical Engineering from IIT Kanpur in India.

Encontre mais informações sobre Ramki Ramakrishna em

Tracks

Segunda-feira, 6 de maio

Terça-feira, 7 de maio

Quarta-feira, 8 de maio