Workshop/Lecture #4 Neuro-Mathematics of Deep Learning

The success of deep learning neural networks is evident, leading to a revolution in the field. The breakthrough is enabled by the discovery of how to train a multi-layered neural net with backpropagation, by the advent of cheap GPU processing power and the availability of huge amounts of data.

The workshop will focus on examples of important convolutional neural networks today and is organized in partnership with SSIMA Re:Imagine Healthcare (Festival of Innovation in MedTech).

About:
We zoom in on an important biomedical application: the large-scale screening for diabetes by automatic analysis of retinal fundus images. With the progression of diabetes, blood vessels begin to leak, and this can be detected at high resolution and at low cost in the retina. We will discuss automated and quantitative biomarkers of early retinal damage to be exploited in deep learning. All algorithms are based on ‘brain-inspired computing’.

We will focus on possible intrinsic mechanisms of deep learning. How does it actually work? Can we devise some mathematical modeling? We can learn a lot from modern brain research, where optical and physiological recording techniques shed new light on how the functional circuits in the brain may be computing: neuro-mathematics.

The biomedical engineer, speaking both languages, is just the right professional to benefit from and contribute to these developments.
The workshop/lecture will be highly visual and is aimed at a broad audience.

Note: it's a bring your own device event, so don't forget it at home.:)
Pre-requisites:

• Mathematica Wolfram installed already. (Kindly use your campus license, if available, or install the 14-days free version: http://support.wolfram.com/kb/12440) - no previous experience required, but if present then it's most welcome. For the first phase, seeing the given code examples working, and playing with it doesn't necessarily need previous Mathematica experience.
• Some knowledge of linear algebra (convolution, eigenvectors), calculus (coordinate systems, vectors) is recommended.
• Existing crash courses by Google (https://developers.google.com/machine-learning/crash-course/) or Udacity (https://eu.udacity.com/course/intro-to-tensorflow-for-deep-learning--ud187) are a plus.

Agenda:
15:30 - 16:00 - Registration & welcome
16:00 - 16:50 - Part I, Introduction to deep learning and convolutional neural nets + 10' QA or small break (if needed)
17:00 - 17:50 - Part II, Back-propagation and face recognition, examples of trained networks + 10' QA or small break (if needed)
18:00 - 18:50 - Part III, A geometric, brain-inspired model for deep learning + 10' QA or small break (if needed)
19:00 - 19:50 - Part IV, Hands-on experience with given Mathematica code
20:00 - 20:30 - Closing & Networking

About the Trainer:
The lecture focused workshop will be held by Prof. Bart ter Haar Romeny from the Eindhoven University of Technology, the Netherlands, Department of Biomedical Technology. His research interests focus on biologically inspired image analysis algorithms, multi-valued 3D visualization, brain connectivity by diffusion tensor imaging, computer-aided diagnosis, and image-guided neurosurgery. He initiated and leads the RetinaCheck project, a large international screening program for diabetes. He published over 250 papers (h-index: 43, > 13500 citations).

Prof. Romeny is President of the Dutch Society for Pattern Recognition and Image Processing and has been President of the Dutch Society for Biophysics & Biomedical Engineering and the Dutch Society of Clinical Physics. He is founder and organizer of SSIMA Re: Imagine Healthcare (the International Summer School of Imaging with Medical Applications).
He is reviewer for many journals, conferences and science foundations, and organized many international Summer Schools. Prof. Romeny is EMBS Distinguished Lecturer, Senior Member of IEEE, Fellow of EAMBES, Board member of IAPR, and Honorary Chair Professor at NTUST, Taiwan.