Dear Deep Learners,
This time our topic is "Putting Deep Learning into Production". Once you have trained and optimized your models, how to put them into production efficiently, to serve your product's and customer's needs?
We have two speakers who have a lot of hands-on experience and will answer this question in detail:
Continuous Integration and Deployment for Machine Learning Applications
Simon Stiebellehner, Head of AI and Bernhard Redl, Data Engineer, craftworks GmbH
Continuous Integration and Deployment has become a standard in modern software engineering environments. How come we see little of these practices applied to machine learning applications yet? The reason is that machine learning is different. Workflow, artifacts, hardware and many other characteristics of ML and resulting applications make traditional CI/CD practices from software engineering unsuitable in significant parts. In this talk, we
- first provide a brief introduction to CI/CD in classic software engineering,
- point out what makes machine learning applications different,
- explain what these differences mean for CI/CD and
- how these principles can be applied to ML applications.
Finally, we provide a practical use case on how we implemented CI/CD practices in one of our deep learning software products.
Computer Vision Models in Production
Jakob Klepp, App Engineer Computer Vision, MoonVision
Your models pass all the tests. Now it's time to run them as fast as possible and maybe on specific hardware. We share our learnings for custom networks developed in PyTorch. Specifically, we'll navigate pitfalls associated with the Open Neural Network Exchange Format (ONNX.ai) and present what's currently working best for industrial use cases.
As usual, the organizers will present hot news and latest papers in Deep Learning. If you would like to contribute to this section, let us know beforehand.
Thanks to Matthias Grabner from WKO for organizing and sponsoring the venue of this meetup edition.
Looking forward to seeing you,
Tom, Rene, Alex, Jan