A0150
Title: Quantifying uncertainty with Bayesian deep learning
Authors: Nadja Klein - Karlsruhe Institute of Technology (Germany) [presenting]
Abstract: Bayesian deep learning fuses deep neural networks with Bayesian techniques to enable uncertainty quantification and enhance the robustness in complex tasks such as image recognition or natural language processing. However, fully Bayesian estimation for neural networks is computationally intensive, requiring the use of approximating inference for virtually all practically relevant problems. Even for partially Bayesian neural networks, there is often a lack of clarity on how to adapt Bayesian principles to deep learning tasks, leaving practitioners overwhelmed by the theoretical aspects, such as choosing appropriate priors. So, how are scalable, reliable, and robust approximate Bayesian methods designed for deep learning? The question is addressed from a methodological perspective with a focus on Bayesian neural networks, Bayesian optimization, and probabilistic programming. Methods are developed that deliver high-accuracy predictions and offer calibrated probabilistic confidence measures in those predictions. This is showcased through examples from different domains and selected open challenges and directions for future research are concluded.
