Bayesian Inference with Neural Networks

Bayesian Inference with Neural Networks Abschlussarbeiten (z. B. Diplomarbeit) Ingolstadt Forschung / Entwicklung I-A-66521 To be filled:   01.06.2020 Work Environment We are a multi-disciplinary team of computer scientists, engineers, physicists, and mathematicians addressing some of today’s biggest challenges in AI, robotics, and autonomous driving technology. Our internal work on perception led to, e.g. the creation of A2D2 (a2d2.audi). You will be part of our team with a culture of openness, collaboration, and community spirit. The highly international nature of our team embodies a belief that creativity thrives when nurtured by diversity of thought, opinion, and experience. Job Purpose/Role Deep learning based approaches (e.g., convolutional neural networks) have produced outstanding results, e.g., in the field of object detection, over the last few years. Most of these results, however, are based on point estimates, e.g., maximum likelihood estimates or – in the best case – maximum a posterior estimates. For safety critical applications such point estimates without proper variance estimations are often insufficient. The assignment of probabilities to the weights of neural networks seems to be the natural solution, but is often numerically not feasible. This master thesis evaluates one novel approach to combine the two worlds (NNs and probabilities) by focusing on a simple Neural Network approach. The master thesis has the following objectives, which should be thoroughly addressed in the thesis: Literature review on related work Deep understanding of Bayesian inference and neural networks Development of a proof of concept algorithm Evaluation of the PoC by application to public available data sets Key Requirements/Skills/Experience Studies in computer science, physics, mathematics, robotics, or equivalent fields Good knowledge and experience in the field of Machine Learning Good knowledge in Bayesian statistics and statistics in general Experience in software development (Python) First Experience in Deep Learning tools (TensorFlow / Pytorch) Good Communication skills High ability to work independently Additional information This position is available at AUDI AG in Ingolstadt ust for reasons of easier readability, only the grammatical male form is used in person names. This always means people of every gender identity. We embrace diversity, actively encourage inclusion and create an environment that fosters each employee’s individuality in the interests of the company. Reference code: I-A-66521