Deep Learning and Adversarial Approach to Time Series Classification and Healthcare Engineering

This dissertation explores the applications of deep learning techniques in time series classification and healthcare, particularly in the context of out-of-hospital cardiac arrest (OHCA). Deep learning algorithms have shown their ability to identify underlying patterns in both fields. The contributions of this dissertation include the development of a multivariate adversarial generator, the use of octave convolutions to improve classification accuracy, a transfer learning algorithm for detecting out-of-distribution data, and a deep learning technique to model OHCA survival and decision-making. These contributions aim to support OHCA community interventions and showcase the potential of deep learning in healthcare. With the availability of large datasets and new sensing technologies, deep learning techniques are rapidly evolving and hold great promise in fields beyond classification tasks.