Repository for Publications and Research Data
News from the ETH Library
Recently Added
Detection of Atmospheric Rivers using GNSS Zenith Wet Delay
Item type: Student Paper
Sarbach, Petra; Hütter, Louis (2026)
Retrieval of Land Surface Parameters from GNSS-IR Data Using Machine Learning
Item type: Master Thesis
Dib, Elissa (2025)
VEXP: A Low-Cost RISC-V ISA Extension for Accelerated Softmax Computation in Transformers
Item type: Conference Paper
Run Wang; Gamze Islamoglu; Andrea Belano; et al. (2025)
On the quantification of robustness and its thresholds
Item type: Journal Article
Beck, André Teófilo; Cao, Alex Sixie (2026)
Structural systems need to be safe enough against foreseeable loads, but they also need to be robust enough against unforeseeable or abnormal loading. In this paper, a novel entropy-based robustness index for arbitrary perturbations is derived for coherent path-dependent systems, which is consistent with information-theoretic and thermodynamic principles. Using a reliability-based robustness index and the entropy-based robustness index, quantitative robustness thresholds are derived that enable the explicit classification of low, medium, and high robustness based on the sensitivity of the system to arbitrary perturbations. Furthermore, relations between the entropy- and reliability- and risk-based robustness indices are explored, where thresholds for the risk-based robustness indices are provided based on the novel entropy-based robustness index. The use of the various robustness indices and the thresholds are exemplified in three case studies, involving a redundant system subjected to various degrees of damage, damage propagation in frame structures, and a network. For the first time, quantitative thresholds for the robustness of coherent path-dependent systems are provided, which can be applied to structures, networks, and more. This paves the way for providing quantitative guidance on acceptable degrees of robustness in such systems, which may lead to more economic and rational systems with an appropriate degree of robustness.
Learning Pareto fronts in high dimensions: How can regularization help?
Item type: Conference Paper
Tobias, Wegel; Filip, Kovačević; Alexandru, Tifrea; et al. (2025)
Modern machine learning methods often have to rely on high-dimensional data that is expensive to label, while unlabeled data is abundant. When the data exhibits low-dimensional structure such as sparsity, conventional regularization techniques are known to improve generalization for a single objective (e.g., prediction risk). However, it is largely unexplored how to leverage this structure in the context of multi-objective learning (MOL) with multiple competing objectives. In this work, we discuss how the application of vanilla regularization approaches can fail, and propose the first MOL estimator that provably yields improved performance in the presence of sparsity and unlabeled data. We demonstrate its effectiveness experimentally for multi-distribution learning and fairness-risk trade-offs.