Overview
Explore the critical aspects of machine learning system reproducibility in this 32-minute conference talk from Data Science Conference Europe 2022. Gain insights into ensuring model performance conclusiveness, system understanding, and error reduction for production deployments - increasingly vital as AI regulations evolve. Learn about key reproducibility components including dataset management, data processing, ML model development, randomness control, hyperparameter tuning, code maintenance, and software environment setup. Discover practical tools and concepts for implementing reproducibility through data versioning, feature stores, metadata management, artifact storage, model registry systems, and containerization approaches. Delivered in-person from Belgrade, this presentation equips practitioners with essential knowledge for building reliable and reproducible machine learning systems.
Syllabus
Reproducibility and versioning of ML systems | Spela Poklukar | DSC Europe 2022
Taught by
Data Science Conference