11-30, 16:35–17:05 (Europe/Amsterdam), Auditorium
Setting up an ML application by creating some ad-hoc dataset and training a model with an unversioned Python script just does not cut it anymore. The MLOps process helps to structure the development and life cycle of any ML application to make sure that data is traceable, and performance is reproducible. In this talk, an automated visual inspection application is used as an example to show a data definition and labeling platform using Django, and an image ingress and streaming service using gRPC. This system can serve as an example for any future ML application as the same principles can be applied. A demonstration of this system is also shown at the end of the session. Having some basic knowledge of Django and gRPC is preferred, but it is not required.
Using deep learning to automate visual inspections in industrial environments is getting more common nowadays. However, if you stick to just training a model with some Python script, storing the artifact somewhere, and creating a webservice, you will quickly find out that you need a better process and infrastructure to guide, manage, and scale machine learning applications. This is where the MLOps process comes in which helps to structure (1) data collection and management, (2) model training experiments, for instance using MLflow, (3) model deployments and services, for instance using Seldon Core, and (4) takes monitoring into account as well.
In this talk, the focus is on the data management part of MLOps using the automated visual inspection application as an example. As an introduction, the application and necessity for MLOps are introduced. After explaining how different steps in the MLOps process can be applied to this application, a closer look is presented on how the definition of the automated inspections and data labeling can be structured using Django. Next to that, an image data ingress and streaming service using gRPC is also shown. Note that the principles given for this specific application apply to any other ML application with different data types as well. The session is closed with a live demonstration of this system.
No previous knowledge expected
My drive is to build robust end-to-end ML systems that can help take efficiency to the next level. In my role as a software architect for over 9 years at Prodrive Technologies, I have gathered experience developing industrial automation applications in C#/.NET, such as a fully automated robotic production line. This includes building software that communicates with both hardware (PLCs, equipment APIs) and other software platforms (MES, PLM). My specialization is in computer/machine vision, where I am responsible for new applications in the industrial automation domain.
In the past 3 years, I have been focusing more on the infrastructure and core SW applications required for ML applications using Python. We are building an AI framework consisting of both open-source and in-house developed code that can structure and automate ML application development (the MLOps process). The main application is automated visual inspection, where we want to provide technicians on the factory floor with explainable deep learning tools to automate the inspection themselves, with minimal need of data scientists and ML or software engineers.
Finally, I am responsible for developing the ML competence both within the company, with workshops, trainings, and pilot applications with students, and outside it with knowledge sharing sessions at conferences, cross-company work groups, and educational institutions like Fontys and Eindhoven University of Technology.