Montonen suggests bridging the divide between data science and machine learning engineering practices, focusing on challenges related to model creation and maintenance.
Key challenges in deploying ML systems include managing training data, model training and evaluation efficiency, measuring production performance, serving predictions, handling data edge cases, re-training, and version control.
Common components in ML systems include a feature store, experiment tracking, model registry, and data quality monitoring, now integrated into many MLOps platforms.
Montonen emphasizes that while tools solve specific system problems, they often overlook external factors shaping how ML systems evolve within a company, emphasizing MLOps as more than just tool usage.
Collection
[
|
...
]