Reproducibility and versioning of ML systems | Spela Poklukar | DSC Europe 2022
The reproducibility of ML systems is an increasingly important topic in the ML community. Reproducibility ensures conclusiveness of the model performance, provides an understanding of how the ML system works and reduces unnecessary errors when the system is deployed into production. With increasing AI regulation, it will soon become a requirement for many ML applications.
In this talk, we will explore different aspects of reproducibility such as reproducibility of the dataset, data processing, ML model, its randomness and hyperparameters, code and SW environment, as well as concepts and practical tools such as data versioning, feature, metadata and artifact store, model registry and containerization that together ensure reproducibility of our experiments.
This speech by Spela Poklukar was held on November 18 at Data Science Conference Europe 2022 in person in Belgrade.
Follow us on social media :
LinkedIn: Data Science Conference
/ 11184830
Instagram: @datasciconf
/ datasciconf
Facebook page: @DataSciConference
/ datasciconference
Website: http://datasciconference.com/
#datascience #ai #ml #artificialintelligence #bigdata #dataanalytics