Deep learning plays a major role in many contemporary artificial intelligence applications. It has shown impressive power in tasks which require human-like perception and understanding of visual or auditory data, or interpretation of vast collections of information. Even if these models show impressive results and are widely used, they are far from perfect and still under active research. This webinar brings up questions and challenges related to deep learning in practice, such as interpretability, trustworthiness, data bias, and quantification of uncertainty. Invited speakers from both industry and academia provide views into challenges in deep learning, and the audience can submit questions for the panel discussion.