Notes
Introduction
- ML-based applications require distinctly new types of software, hardware and engineering system.
- e.g collecting, preprocessing, labeling, reshaping dataset rathen writing code and also deployed in different way e.g.
- specialize hardware
- quality assurance method
- end to end workflow
Why Now? The Rise of Full Stack Bottlenecks in ML
- Deployment Concerns
- robustness to adversarial influences, spurious factor
- privacy, security
- Cost
- Resource for training
- Annotation
- Latency, Power
- Accessibility
- It should be everyone can use not just PhD-level
MLSys: Building a New Conference at the Intersection of Systems + Machine Learning
- MLSys conference
Conclusions
- Many things to learn e.g. software, hardware, monitoring.
Resources
- https://github.com/HuaizhengZhang/Awesome-System-for-Machine-Learning/blob/master/paper/mlsys-whitepaper.pdf