Adversarial Attacks and where to find them

Barbara Hammer, University of Bielefeld

Adversarial attacks refer to marginal changes of inputs for popular machine learning models such as deep networks, which unexpectedly change the output classification. As an example, changing only a few pixels of a traffic sign image can change a stop sign to a perceived speed limit. Attacks can happen even in the physical worls and if the deep network itself is unkown. In the lexture, the focus will be on the question, what makes an adversarial attack, how to actively create those, and how to measure vulnerability with respect to adversarial attacks. How to find robust features, which are not vulnerable to adversarial attacks, and how to intuitively inspect the data topology in attacked regions and robust mitigations thereof.

Making robots intelligent through vision

Javier Gonzalez-Jimenez, University of Malaga
An intelligent mobile robot is sensorimotor machine that autonomously operates in an unknown, unpredicted world. To achieve such capability, the robot is equipped with a variety of sensors to perceive and understand its surroundings, being vision one of the most powerful and affordable sensing modalities for that. This talk will cover the main issues in robotics vision, including obstacle detection, localization and semantic mapping. Particularly, we will review the challenging visual SLAM problem which tries to build a map of the environment while the robot estimates its pose (position and orientation) from a sequence of images provided by a camera onboard.

Biomedical Imaging: From Information Processing to Galisonian Perspective

Xiaoyi Jiang, University of Muenster
Imaging has become an indispensable and powerful tool in biology and medicine for basic research and clinical practice. The specific image characteristics and problems in these fields have motivated researchers to develop novel concepts and algorithms. This talk emphasizes the fundamental research view of biomedical imaging and discusses a number of challenges and related concepts and algorithms. The imposing development in biomedical imaging not only enables numerous applications, but also provides a driving force for life sciences from a Galisonian perspective.