Carl Zeiss AG – The Road to MLOps: Machine Learning development at ZEISS

At Carl Zeiss AG, our goal is to offer responsible Machine Learning (ML/AI) applications our customers can trust. To build trust in ML/AI, an end-to-end solution to operate Machine Learning is imperative.[1]
In this talk, we will show how our team is set up to develop sophisticated state-of-the-art ML/AI applications. The success of these applications depends on the capability to manage the full ML/AI development and life-cycle management through a professional MLOps setup.[2] We will outline the path to MLOps and how it depends on selecting the right people, processes, technologies, and operating models. We will share our journey towards sustainable ML/AI applications based on real ML/AI use cases at Carl Zeiss AG. To build sustainable ML/AI products that deliver real value, use cases need to go beyond proof-of-concept status. Thus, the need to follow a fully functional development, operationalization, and automation cycle. We believe in the power of standardization at the core and flexibility at the shell to develop reliable and reproducible ML/AI applications. MLOps drives this through the entire lifecycle of ML/AI models, from design to implementation to management. We will introduce the challenge of dealing with data and concept drift and how to deal with it in our MLOps setup. Our goal is to develop all necessary building blocks of a state-of-the-art MLOps solution to ensure fairness, accountability, and transparency.[3]

[1] “ML Integrity: Four Production Pillars For Trustworthy AI” by Nisha Talagala Forbes Jan. 2019

[2] “Datasheets for Datasets” Kate Crawford et al.

[3] Keeping an eye on AI with Dr. Kate Crawford Microsoft Research Feb, 2018

Speaker

Dr. Lydia Nemec, Manager, Head of AI Accelerator | Carl Zeiss AG

As Head of the ZEISS AI Accelerator, Dr. Lydia Nemec and her team work closely together with all ZEISS businesses to create fully operational machine learning products that drive business success for ZEISS. Coming from academic research in high-performance computing and numeric algorithm development, Lydia started her exciting career path in theoretical and computational physics as a post-doc at Fritz Haber Insitute of the Max Planck Society and TU Munich. Before joining ZEISS, Lydia worked as a Senior Data Scientist on Machine Learning development and cloud-based data products at Knorr-Bremse. As part of the ZEISS Data & Analytics unit, Lydia leads the ZEISS Data Science team and is at the forefront of ZEISS´ next generation of AI-infused product and service innovations.