This article discusses the benefits of a streaming-first infrastructure for real-time machine learning, including:
- fast responses,
- continual learning,
- and adapting to changes in data distributions in production.
The article also explores the advantages of request-driven event-driven architecture and monitoring for continual learning. The iteration cycle should be order minutes, and there are great cases for continual learning, including recommendation systems. Finally, the article discusses barriers to streaming-first infrastructure and how it can be the future of real-time machine learning.
