T01: Quantum Machine Learning - Hype of Heuristic
OTH Regensburg, Prof. Dr. Wolfgang Mauerer und Team
Quantum machine learning (QML) is an emerging technique that leverages the principles of quantum mechanics for potential computational speed-ups. It has been shown that certain problems can be solved more efficiently using quantum algorithms over classical approaches. However, the practical utility of these algorithms is limited on the current generation of quantum computers, so-called noisy intermediate-scale quantum (NISQ) systems, as they only offer a limited amount of qubits and are prone to noise and imperfections that strongly limit possible circuits depth and thus the length of quantum computations. Hybrid variational algorithms are considered key candidates for exploiting advantages of near-term quantum devices, but could also be beneficial in post-NISQ systems because of their resource efficiency.
In this tutorial, we discuss how within the class of hybrid variational algorithms, quantum machine learning (QML) has shown promise by moving certain parts of classical machine learning to quantum computers. QCs will, despite common misperceptions, likely be inapt for handling large amounts of data. Consequently, we focus on qantum reinforcement learning (QRL), which requires little training data. However, we also discuss how QML has been shown to outperform classical machine learning for certain tasks, and how this could benefit ML tasks by, for instance, reducing the number of parameters, or in terms of energy efficiency.
T02: Kaapana Demo - HandsOn
DKFZ Heidelberg, Prof. Dr. Klaus Maier-Hein, Ünal Akünal und Team
T03: Hands-on Tutorial on Implicit Neural Representations (INR) in Medical Imaging
Universität zu Lübeck, Prof. Dr. Heinrich Mattias, Ziad Al-Haj Hemidi, Fenja Falta, Christoph Grossbroehmer
Complementary to the recent success in population-based training of deep neural networks for medical image analysis tasks there has been an exiting development and advancement of optimisation driven approaches using Implicit Neural Representations (INR). INRs, which model e.g. mappings between input coordinates and output gray values as continuous functions, are promising in various vision tasks including motion estimation, image reconstruction, denoising, superresolution, and compression. Over the last few years, those ideas have been transferred into the medical imaging field, where they offer unique advantages in adapting to task- and instance-specific challenges that have thus far prevented a wider spread adoption. Our tutorial will provide both a comprehensive overview lecture of INR research in the medical imaging domain as well as practical demonstrations and hands-on learning for interested students and researchers with little or no prior knowledge of the topic.