Title: '''Self-learning Software for Predictive Systems in Autonomous Driving'''

'''Abstract:'''

The next steps towards human-like AI autonomous driven cars are paved by the development of self-adapting software systems that can reason beyond traditional deep learning perception methods. In this talk, an insight into current AI activities undergoing within Elektrobit will be given. This includes ''Deep Grid Net'', a deep neural network designed for inferring context awareness from grid data, Elektrobit’s approach to learning how to drive in a simulator using Deep Reinforcement Learning, as well as our ''Generative One-shot Learning'' framework built to learn representations of data patterns from single object instances.