Christian Kästner
Associate Professor, School of Computer Science, Carnegie Mellon University
Invited Talk: Toward a system-wide and interdisciplinary perspective on ML system performance
Slides
Video
Abstract:
Building production systems with machine learning components requires a system-wide design and analysis process that involves many ML and non-ML components and many stakeholders from different backgrounds. Performance concerns need to be balanced with other qualities and design goals, such as update frequency, telemetry needs, and deployment and privacy concerns. It requires negotiation between data scientists, software engineers, operators, and other stakeholders to agree on constraints, assumptions, and system structure. In this talk, I’ll emphasize the system nature of production systems with ML components and borrow from the software architecture view to analyse and design systems that meet performance and other quality goals.
Biography:
Christian Kästner is an associate professor in the School of Computer Science at Carnegie Mellon University, interested in limits of modularity and complexity caused by variability in software systems. I develop mechanisms, languages, and tools to implement variability in a disciplined way despite imperfect modularity, to understand feature interactions and interoperability issues, to detect errors, to help with nonmodular changes, and to improve program comprehension in software systems, typically systems with a high amount of variability. Among others, I have developed approaches to parse and type check all compile-time configurations of the Linux kernel in the TypeChef project.
Dr Kästner received his Dr.-Ing. in Computer Science from Otto-von-Guericke University, Magdeburg, Germany in 2010. After serving 2 years as a Research Assistant at the University of Magdeburg, he came to CMU in 2012, initially as an Assistant Professor.