Computer Vision Deep Learning Development GPU Machine-LearningSee in schedule: Fri, Jul 30, 14:15-15:00 CEST (45 min) Download/View Slides
An important, but often overlooked aspect of developing a high-quality deep learning model is the iteration speed.
If you can iterate faster, you can try out more ideas and over time you get better results.
In this talk, you will learn about the different tricks you can use to train a great machine learning model in a shorter amount of time.
In particular, I will discuss how we optimized our deep learning development life cycle at Mindpeak to create robust deep learning models for cancer diagnostics that work in vastly different laboratory settings.
The goal of this talk is to point to the most important aspects which you can adjust to speed up the time it takes to go from idea to validated result.
I will talk about many different aspects like task prioritization, data processing, communication, GPU parallelization, code quality, unit tests, continuous integration, data fit and profiling for speed.
So hopefully, after the talk, you should be able to point to some items that you could do to improve the iteration speed when developing machine learning models.
There are no strict requirements for the talk, but you probably obtain the highest benefit if you have gained some initial experience in developing machine learning models.
Type: Talk (45 mins); Python level: Beginner; Domain level: Intermediate
Marc has a background in computer science and computational neuroscience and is currently the CIO at Mindpeak.
At Mindpeak, he creates machine learning solutions to help cancer experts with better diagnostic tools.
Before that, he founded and grew the Lemundo GmbH as the company's CTO to more than 25 people.
He has given several talks on conferences and meetups in the past and is an avid learner who loves challenging problems. Since he has learned from so many other people, he likes to pass on his experiences in his personal blog at paepper.com/blog.