Rec. ITU-T Y.3531 (09/2020) Cloud computing – Functional requirements for machine learning as a service Summary History FOREWORD Table of Contents 1 Scope 2 References 3 Definitions 3.1 Terms defined elsewhere 3.2 Terms defined in this Recommendation 4 Abbreviations and acronyms 5 Conventions 6 Overview of machine learning 6.1 Introduction to machine learning 6.2 Generic process of machine learning 6.3 Machine learning ecosystem 6.3.1 Data provider 6.3.1.1 Data supplier 6.3.1.2 ML data provider 6.3.2 ML model provider 6.3.3 ML framework provider 6.3.4 ML framework customer 7 Machine learning as a service 7.1 System context of MLaaS 7.2 CSN:machine learning data provider 7.2.1 Data labelling provision 7.3 CSN:machine learning model developer 7.3.1 ML model development 7.3.2 ML model registration 7.4 CSP:machine learning service provider 7.4.1 ML data audit 7.4.2 Data feature engineering 7.4.3 ML model training 7.4.4 ML model testing 7.4.5 ML model training monitoring and reporting 7.4.6 ML model deployment management 7.4.7 Retraining policy management 7.5 CSC:machine learning service user 7.5.1 ML service use 8 Functional requirements of MLaaS 8.1 ML data collection and storage requirements 8.2 ML data labelling requirements 8.3 ML data pre-processing requirements 8.4 ML data analysis and feature engineering requirements 8.5 ML model training requirements 8.6 ML model monitoring requirements 8.7 Trained ML model deployment and retraining requirements 9 Security considerations Appendix I Use case of MLaaS for operation perspectives I.1 ML data annotation and labelling management I.2 Model training with user configuration I.3 Report learning result and re-training ML model I.4 Distributed training with multiple worker nodes I.5 Model testing and optimizing the model quality includes hyperparameter tuning I.6 Model monitoring to issue alerts of abnormal or unsuspected learning process I.7 Model deployment and monitoring I.8 Automated machine learning in cloud computing Appendix II Use case of MLaaS for application perspectives II.1 Object recognition model development in the cloud computing environment II.2 Traffic speed prediction and monitoring service II.3 Image recognition II.4 Face recognition II.5 Image segmentation model development II.6 Generative adversarial model development Bibliography