- How should software systems be designed to support the full machine learning lifecycle, from program- ming interfaces and data preprocessing to output interpretation, debugging and monitoring?
- How should hardware systems be designed for machine learning?
- How should machine learning systems be designed to satisfy metrics beyond predictive accuracy, such as
power and memory efficiency, accessibility, cost, latency, privacy, security, fairness, and interpretability?
- high-level systems for ML that support interfaces and
workflows for ML development—the analogue of traditional work on programming languages and software
- low-level systems for ML that involve hardware or software—and that often blur the lines
between the two—to support training and execution of models, the analogue of traditional work on compilers and