Enhancing Uncertainty Estimation of Multimodal Deep Learning

This thesis presents novel methods for accurately measuring the uncertainty in predictions made by multimodal deep learning models. These models not only understand various data types, such as images and text, but also provide reliable uncertainty estimation. This allows the effective use of diverse data sources in real-world applications and makes deep learning models more trustworthy and reliable. The proposed methods have significant potential in safety-critical domains like medical diagnosis and autonomous driving, where integrating multiple data sources and ensuring model reliability are essential.