Speaker
Description
A software suite to prepare (CMS) Open Data for machine learning purposes is introduced. In this presentation, different approaches and their suitability to extract low-level information will be compared. The full chain is implemented with the help of high-performance computing infrastructures, and a study of available data formats and data tiers is conducted. As a proof-of-concept, the work of an exchange summer student shows how quickly learners can get started with data exploration and feature interpretation without prior experience. Key takeaways from a teacher’s or supervisor’s perspective and outreach potential are reviewed.
Aiming for an experiment-independent framework to study safety and robustness of typical deep-learned algorithms, the concept of a wrapper around different applications is introduced. Within this infrastructure, researchers can implement a variety of so called adversarial attacks and defenses, which is demonstrated with examples.
