Courses

Back

A8 - Front-End Vision and Deep Learning 2018

From: November 21 to December 20, 2018


 

 

Goal of this course

Deep learning, self-organization and plasticity, convolutional neural networks, geometry engine, contextual Gestalt processing: the field known as 'brain-inspired computing' is one of the most promising avenues in medical image computing today. The results are spectacular, and although it is clear that brain mechanisms and deep learning are related, both are still largely black boxes.

 

To make new breakthroughs in this spectacular arena the interplay between the fields of brain imaging and physiology, neural network informatics and fundamental mathematics is essential. However, these fields speak different languages, and interaction is nontrivial.

 

This course focuses on exactly that: we will discuss in detail modern findings in the neurophysiology, connectivity and functionality of the visual system, the best studied brain function today. And we deeply look into the mathematical background of Deep Learning. The goal is to develop and better understand both highly effective medical computer-aided diagnosis systems, as well as modern models for visual perception.

 

We give insight in modern approaches to deep convolutional neural nets, and we will teach a number of well-established mathematical modeling techniques in detail, in particular multi-scale and multi-orientation differential geometry, geometric reasoning, models for self-organization and plasticity, and geometric neural feedback, leading to effective adaptive operations. We present the theory in an axiomatic, intuitive and fundamentally understood way.

 

We discuss modern brain imaging methods at all scales, and innovative models for human and computational vision, such as the role of retinal stellate amacrine cells, on-off channels, colorization and temporal processing. The course is interspersed with working and powerful applications in medical image analysis, such as computer-aided detection of breast tumors, polyp detection in 3D virtual colonoscopy, deblurring, invariant feature detection, adaptive geometry-driven diffusion, development of retinal vessel biomarkers, and contextual Gestalt-based operators to deal with missing data.

 

This is an intensive course, 4 weeks every Wednesday and Thursday, starting 21 November 2018.
Each morning of 3 lectures is followed by an afternoon computer lab (all software code is supplied). We exercise with all developed notions, exploiting the high-level 'play and design' functionality of Wolfram’s Mathematica 11.

 

 For more detailed information please see: www.frontendvision.net/2018/

Back

 ASCI Tweets