Deep Learning Series - Session 3: Deep Dive - Designing Experiments

Overview

In the first part, we will take a deeper dive into designing, training, and tuning deep learning models. We will show how MATLAB’s deep learning apps can help you edit neural networks, and devise and run experiments. We will also show how to leverage cloud compute resources to speed up your training.

In the second part, Arnie Berlin gives a step-by-step explanation of the training workflow developed for research on head and neck tumor segmentation from 7 modality/channel MRI images. Semantic segmentation training on Multi-modal MRI datasets can be very time consuming, even on single-GPU computers. Scalability from local CPU-only and single-GPU computers for troubleshooting and refinement to cloud based, high performance, multi-GPU computers is desired to minimize the more extensive time required for leave-one-out training. Development of the workflow was a collaboration between the University of Freiburg Medical Research Center (UKLFR) and MathWorks.

Highlights

• Using the Deep Network Designer app to graphically create, edit, and train models
• Tracking and running modeling runs with the Experiment Manager app for rapid, automated iteration
• Introducing the extended deep learning framework to customize and train advanced neural networks

About the Presenters

Arnie Berlin has been a Senior Application Engineer at MathWorks for 7 years. With 29 years prior experience in the automated inspection industry, He works with customers on a broad range of computer vision applications. He currently specializes in classical machine and deep learning as well as GPU and cloud-based workflows.

Christoph Kammer is an application engineer at MathWorks. He supports customers in many different industries in the areas of machine and deep learning, image and signal processing and deployment to embedded or enterprise systems. Christoph has a master’s degree in mechanical engineering from ETHZ and a PhD in Electrical Engineering from EPFL, where he specialized in optimization and control design as well as the control and modelling of power systems.

Toon Weyens is an application engineer at MathWorks in Eindhoven, Netherlands. He supports innovative companies in automation and machinery, automotive, and aerospace industries by helping them use MathWorks software to analyze data, develop algorithms, create mathematical models, and scale to run on clusters, GPUs, and clouds. Prior to joining MathWorks, Toon was a postdoctoral researcher at the ITER Organization. He holds a M.Sc. degree in Energy Engineering from the University of Leuven, a M.Sc. degree in Nuclear Fusion Science and Technology from the Universidad Carlos III in Madrid, and a Ph.D. degree in Applied Physics from Eindhoven University of Technology.