Bøger udgivet af TUDpress

In an era of complex deep learning architectures like transformers, CNNs, and LSTM cells, the challenge persists: the hunger for labeled data and high energy. This dissertation explores Echo State Network (ESN), an RNN variant. ESN's efficiency in linear regression training and simplicity suggest pathways to resource-efficient, adaptable deep learning. Systematically deconstructing ESN architecture into flexible modules, it introduces basic ESN models with random weights and efficient deterministic ESN models as baselines. Diverse unsupervised pre-training methods for ESN components are evaluated against these baselines. Rigorous benchmarking across datasets - time-series classification, audio recognition - shows competitive performance of ESN models with state-of-the-art approaches. Identified nuanced use cases guiding model preferences and limitations in training methods highlight the importance of proposed ESN models in bridging reservoir computing and deep learning.

This work contributes to the field of numerical simulation of multiphase flows with a focus on gas bubbles in liquids. The immersed boundary code PRIME allows the explicit capture of phase boundaries of the solid-liquid and solid- gaseous type. The present work contributes to a third boundary type, the interface between gas and liquid as it appears in bubbly flows. This is a challenge due to the deformability of the bubbles. The basic idea is two-fold: first, the interface motion and the interfacial forces are derived from the Navier-Stokes equations, allowing a Lagrangian specification of the surface motion and the surface loads. Second, based on these equations of motion, the bubble shape is captured via a non-local parametric representation. Thus, the bubble is explicitly known as a continuous object. This is an essential difference to conventional methods, where it is common to define the surface position via marker points or, implicitly, by an indicator function. Through this continuous, Lagrangian representation, the bubble can change position and shape and is coupled locally with the fluid field.