AI Engineering & MLOps het bouwen van betrouwbare AI-systemen Petra Heck – Sep 2022

Petra Heck – Fontys Hogeschool ICT • M.Sc. Computer Science, Software engineer & quality consultant • Lecturer Software Engineering since 2012 • PhD Computer Science (Quality of agile software requirements) • Lectoraat AI & Big Data since 2016 – postdoc AI engineering • Kenniscentrum AI for Society since 2022 – senior researcher – Quality model for trustworthy AI systems – Tools, techniques and frameworks for building trustworthy AI systems – AI for health projects https://fontysblogt.nl/author/petraheck/

From DevOps to MLOps

Definitions AI autonomous machine intelligence Machine Learning algorithms to build AI Deep Learning machine learning with neural networks https://livebook.manning.com/book/deep-learning-with-javascript/chapter-1/v-3/

Machine Learning Applications https://livebook.manning.com/book/deep-learning-with-javascript/chapter-1/v-3/ End User Application Service

From DevOps to MLOps MLOps = ModelOps = AIOps = AI Engineering = ML engineering = … https://fontysblogt.nl/ai-engineering-and-mlops/

MLOps: building production-ready ML systems Production-ready ML systems should: • be developed with a collaborative team across the full machine learning lifecycle; • deliver reproducible and traceable results; • be continuously monitored and improved. (mlops.community/manifesto/)

[CH1] Elicitation of Data and Model Requirements [CH2] Modularizing the Application [CH3] Design through Experimentation [CH4] Data and Model Management [CH5] Testing Heck, Petra, Gerard Schouten, and Luís Cruz. "A Software Engineering Perspective on Building Production-Ready Machine Learning Systems." Handbook of Research on Applied Data Science and Artificial Intelligence in Business and Industry. IGI Global, 2021. 23-54. Building Production-Ready ML Systems https://fontysblogt.nl/a-toolbox-for-the-applied-ai-engineer/

MLOps Open Source Tools Explaining predictions & models Privacy preserving ML Model & data versioning Model training orchestration Model serving and monitoring Neural architecture search Reproducible notebooks Visualization frameworks Industry-strength NLP Data pipelines & ETL Data labelling Data storage Functions as a service Computation distribution Model serialization Optimized calculation frameworks Data stream processing Outlier and anomaly detection Feature engineering Feature stores Adversarial robustness Categories of open-source tool support for production ML, adapted from (EthicalML, 2020)

Cases

AI Engineering Education @ Fontys • Each semester project from external organization • Hands-on applied machine learning, “no math” • Covers full machine learning life cycle • Combines software engineering and machine learning • Includes data engineering and data visualization Turning Software Engineers into AI Engineers

Conclusion

Quality Model for Trustworthy AI Systems https://fontysblogt.nl/a-quality-model-for-trustworthy-ai-systems/

AI engineering @ Fontys - Future • Update toolbox for trustworthy AI systems • Student projects at/with ICT organizations/departments • Partners for long-term innovation or research project [email protected] https://www.linkedin.com/company/fontys-kenniscentrum-applied-ai-for-society