Svetlana Pavlitskaya (M.Sc.)

Wissenschaftliche Mitarbeiterin

Werdegang

Svetlana Pavlitskaya studierte von 2011 bis 2018 Informatik an der RWTH Aachen. Die Schwerpunkte ihres Studiums lagen in den Gebieten Deep Learning und Computer Vision. Ihre Masterarbeit verfasste Sie über die Integration von Deep-Learning Komponenten in autonome Fahrzeugarchitekturen.

Seit November 2018 ist Svetlana Pavlitskaya als wissenschaftliche Mitarbeiterin am Forschungszentrum Informatik (FZI) in der Abteilung Technisch Kognitive Systeme (TKS) angestellt.

Publikationen

Konferenzbeitrag (3)

1. Adversarial Black-Box Attacks on Vision-Based Deep Reinforcement Learning Robotic AgentsDetails

   Atanas Tanev, Svetlana Pavlitskaya, Joan Sigloch, Arne Roennau, Ruediger Dillmann, J. Marius Zöllner, 2021

2. Using Mixture of Expert Models to Gain Insights Into Semantic SegmentationInfoDetails

   Svetlana Pavlitskaya, Christian Hubschneider, Michael Weber, Ruby Moritz, Fabian Hüger, Peter Schlicht, J. Marius Zöllner, 2020

   Not only correct scene understanding, but also abilityto understand the decision making process of neural net-works is essential for safe autonomous driving. Currentwork mainly focuses on uncertainty measures, often basedon Monte Carlo dropout, to gain at least some insight intoa models confidence. We investigate a mixture of expertsarchitecture to achieve additional interpretability while re-taining comparable result quality.By being able to use both the overall model output aswell as retaining the possibility to take into account individ-ual expert outputs, the agreement or disagreement betweenthose individual outputs can be used to gain insights into thedecision process. Expert networks are trained by splittingthe input data into semantic subsets,e.g. corresponding todifferent driving scenarios, to become experts in those do-mains. An additional gating network that is also trained onthe same input data is consequently used to weight the out-put of individual experts. We evaluate this mixture of expertsetup on the A2D2 dataset and achieve similar results to abaseline FRRN network trained on all available data, whilegetting additional information.

3. Feasibility and Suppression of Adversarial Patch Attacks on End-to-End Vehicle ControlInfoDetails

   Svetlana Pavlitskaya, Sefa Ünver, J. Marius Zöllner, 2020

   In an end-to-end vehicle control scenario, where a deep neural network is trained on visual input solely, adversarial vulnerability leaves a possibility to manipulate the steering predictions. Patch-based adversarial attacks present an especially serious menace, because they can be performed in the real world by printing out a generated universal pattern. However, the boundary conditions and feasibility of such attacks to compromise the security of autonomous vehicles have been only sparsely studied so far.We demonstrate and evaluate such attacks in the CARLA simulative environment under different weather and lighting settings, while conducting experiments in open and closed loop attack scenarios. Our findings reveal that attack strength is highly dependent on the surrounding location as well as on environment conditions. We also observe that attack success in an open loop scenario only partially coincides with that in a closed loop scenario. This analysis helps to set the stage for future experiments on public roads.Furthermore, we propose a defense concept to remove malignant perturbations from an input image, which does not affect its salient regions. We analyze deviations from the unattacked vehicle trajectory both on adversarial and suppressed inputs.

Export Suchergebnis .bib

Kontakt

Telefon: +49 721 9654-374
E-Mail: pavlitsk@dont-want-spam.fzi.de

Zurück