Phd Student Ai

We are searching for a research scientist (M/F/diverse) to join our artificial intelligence (AI) team in Berlin. As part of the central pre-development department, our goal is to develop and enable the use of AI in future Continental products. You will work in an innovative team of specialists and students with the goal of a Ph.D. The traceability of an AI function's decision is essential, especially for safety-critical systems such as autonomous driving. In this context, two aspects will be considered in this thesis. First, the development of methods for the plausibility AI perception function predictions by estimating uncertainties in the decision through additional input of knowledge. And second, the investigation of methods for self-explanation of predictions by extracted knowledge from the AI function with the goal to realize an assisted error correction. The goal is to develop a self-explanation component, including a plausibility check of predictions of the AI perception function, which allows an interpretation of the reliability by means of KPIs (Key Performance Indicators), as well as showing strategies how errors in the decisions can be corrected. Example: As an example, a system with a perception function and a subsequent path planning is considered. The perception function is a deep neural network (DNN) that determines a semantic segmentation of the input image data. The plausibility check of the prediction is intended to determine the uncertainty in the decision, such that this can be taken into account during path planning. However, since the confidence statements of neural networks are not always trustworthy, further methods must be applied. It is possible to increase the trustworthiness of the prediction by calibrating neural networks. However, knowledge, e.g., physical correlations, can also be used to make the prediction more plausible. As an example, motion models can be used to check whether the motion resulting from the detections is plausible. In addition, probabilistic methods can be used. For the self-explanation of AI function predictions, among other things, the internal decisions in DNNs are used. For this purpose, knowledge is extracted from a DNN in the first step and finally used in the self-explanation component to increase the comprehensibility of the decision. For example, in the case of pedestrian detection, the first step could be to determine, which concepts (e.g., hands, torso, or head) play a role in the decision and later use these concepts to explain why a pedestrian was detected. This case becomes interesting if, for example, a person is occluded and only certain concepts are visible. Finally, detected failure cases can be analyzed using the self-explanation and appropriate strategies can be developed to fix them. For example, the used training data might contain too few or hardly any persons with occluded hands. Goals of thesis: - Development of approaches for the plausibility check of AI perception functions decisions, by means of estimation of uncertainties using additional world knowledge and extracted concepts - Development of a self-explanation component for explaining DNN decisions with the help of extracted concepts from the DNN - Creation of strategies to resolve failure cases using the self-explanation component - Integrating the plausibility check and self-explanation component into a simulation to run experiments and demonstrations - Development of KPIs to check the effectiveness of the developed methods as well as to monitor the AI function during runtime with respect to plausible and uncertain predictions Working in the AI Campus in Berlin means being part of a highly motivated and growing team of researchers working in the field of artificial intelligence. Being located in one of the German centers for AI, we strive to work on the frontier of research by having open exchange with experts and highly influential players in the community. Frequent talks, gatherings, and discussions as well as formal and informal team events ensure prompt spread of information. In our freshly established team of mostly Ph.D. students, we are maintaining an harmonious atmosphere of easy-going collaboration and exchange. Despite this inviting culture of togetherness, we will provide you with enough flexibility in terms of working time and location. Doing a Ph.D. in Continental is going to equip you with exceptional qualification, network, and industrial experience in the future field of autonomous driving. We are looking forward to meet you at the campus **What we offer** - In addition to the interesting field of activity of the function, the city of Berlin offers a high recreational value and stands for quality of life - You will work in an innovative work environment, namely the "co-working space AI Campus" - Become part of our motivated team - we are looking forward to you