Principles and Practices of Building Machine Learning Systems
Explore the challenges and best practices of building machine learning systems in production environments in this conference talk from NDC Oslo 2023. Delve into why many ML projects fail despite increasing adoption, and understand the unique complexities of working with models as artifacts. Learn about the specialized infrastructure required for monitoring training and inference performance. Examine the proliferation of tools and techniques aimed at simplifying model building and management, and critically assess their effectiveness. Gain insights into the key differences between ML systems and traditional software engineering, including challenges posed by large datasets, extended feedback loops, non-deterministic behavior, and GPU acceleration requirements. Benefit from a detailed case study of large-scale recommender systems implementation at Storytel, offering practical lessons for building robust ML systems in real-world scenarios.