Repository for Publications and Research Data
News from the ETH Library
Recently Added
Across Hosts and Nations: Genomic Insights into Drivers of Viral Transmission
Item type: Doctoral Thesis
Valenzuela Agüí, Cecilia (2026)
The spread of infectious diseases is shaped by heterogeneity among hosts and viral lineages, and by the epidemiological and ecological contexts in which transmission occurs. Understanding these sources of variation is especially important in emerging epidemics and cross-species transmission systems, where incomplete surveillance and complex transmission processes limit direct observation. In this thesis, I use viral genomic data and birth-death phylodynamic models to investigate how transmission heterogeneity shapes epidemic dynamics, using SARS-CoV-2 as the main empirical system. This thesis combines methodological development with empirical applications. First, I extend the multi-type birth-death framework to incorporate predictor-driven formulations, enabling external information to be integrated directly and formally tested, and apply this approach to the early spread of SARS-CoV-2 in Europe. Second, I present a birth-death model with non-inherited host types to study host-related variation in infectiousness, focusing on asymptomatic transmission. Applied to a Swiss screening-based SARS-CoV-2 genomic dataset, this model estimates both the proportion of asymptomatic infections and their relative infectiousness. Third, I analyse genomic data from companion animals and their owners to investigate cross-species transmission. These analyses support repeated in-household transmission from humans to companion animals, with little evidence for sustained onward transmission among companion animals, and suggest variant-specific differences in susceptibility in animal hosts. Overall, the results show that biologically informed phylodynamic models can improve the interpretability of transmission inference while highlighting the importance of aligning model complexity with the available data and epidemiological context. More broadly, this thesis demonstrates the potential of multi-type birth-death models to study transmission heterogeneity across scales within a One Health framework.
Test-time Offline Reinforcement Learning on Goal-related Experience
Item type: Conference Paper
Bagatella, Marco; Albaba, Mert; Hübotter, Jonas; et al. (2026)
Foundation models compress a large amount of information in a single, large neural network, which can then be queried for individual tasks. There are strong parallels between this widespread framework and offline goal-conditioned reinforcement learning algorithms: a universal value function is trained on a large number of goals, and the policy is evaluated on a single goal in each test episode. Extensive research in foundation models has shown that performance can be substantially improved through test-time training, specializing the model to the current goal. We find similarly that test-time offline reinforcement learning on experience related to the test goal can lead to substantially better policies at modest compute costs. We propose a novel self-supervised data selection criterion, which selects transitions from an offline dataset according to their relevance to the current state and quality with respect to the evaluation goal. We demonstrate across a wide range of high-dimensional loco-navigation and manipulation tasks that fine-tuning a policy on the selected data for a few gradient steps leads to significant performance gains over standard offline pre-training. Our goal-conditioned test-time training (GC-TTT) algorithm applies this routine in a receding-horizon fashion during evaluation, adapting the policy to the current trajectory as it is being rolled out. Finally, we study compute allocation at inference, demonstrating that, at comparable costs, GC-TTT induces performance gains that are not achievable by scaling model size.
Forecasting Routing Inputs
Item type: Conference Paper
Mégret, Jean Bernard David; Schneider, Tibor; Vanbever, Laurent (2026)
BGP messages exchanged over peering sessions in the Internet are currently believed to be unpredictable. Today, network management systems (e.g., for verification, analysis, and optimisation) manage this unpredictability in one of two ways: They either consider future routing inputs in a case-by-case manner, or they consider all possible ones, including highly unlikely scenarios. Thus, the impact or risk of routing inputs on the network is either under- or overestimated. In this paper, we present a probabilistic perspective to bridge the gap between these two extremes. We explore the idea of building forecasting models of routing inputs and begin to answer the questions of how feasible it is and what it is useful for. This view paves the way for a large class of new systems trying to provide guarantees in spite of uncertainty.
Coupled Hyperbolic Subsurface Flow (CHSF) Model to Solve Buoyant Two-Phase Flow and Transport in Heterogeneous Porous Media
Item type: Journal Article
Armin Damon Riess; Rasim Hasanzade; Hamdi A. Tchelepi; et al. (2026)
Traditional methods for simulating buoyant two-phase flow and transport in porous media involve solving large systems of tightly coupled discrete equations, which involves non-local operations. In [1], a new approach was proposed which requires no global system to be solved and thus only relies on local computations. This makes it well-suited for large-scale parallel processing on multiple CPUs and GPUs. The new coupled hyperbolic subsurface flow (CHSF) model for two phase flow and transport, was compared to the Implicit Method and an established general purpose reservoir simulator (GPRS), showing that it is highly efficient and accurate for homogeneous media. However, numerical artifacts appear if it is employed for heterogeneous media. In this work, these artifacts are addressed, and to resolve the deficiencies, a novel multi-dimensional Rankine-Hugoniot-Riemann (MD-RHR) solver is devised. Numerical experiments demonstrate that with this MD-RHR solver the CHSF model can deal with heterogeneous permeability fields and accurately reproduces reference results.
[1] Jenny, P., Hasanzade, R., & Tchelepi, H. (2023). Tightly coupled hyperbolic treatment of buoyant two-phase flow and transport in porous media. Journal of Computational Physics, 489, 112–205. https://doi.org/10.1016/j.jcp.2023.112205
Why Do We Suffer for Fun? Ordeal Pleasure in Souls-like Games
Item type: Conference Paper
Fan, Xiaofeng (2026)
Souls-like games turn repeated failure into a distinctive form of satisfaction. This paper theorizes that process as ordeal pleasure: the transformation of severe but trusted difficulty into retrospective and prospective meaning. The Ordeal Pleasure Framework (OPF) identifies a core dyad and a social amplifier. Ludic Cultivation makes failure usable by linking fair, learnable adversity to mastery. Aspirational Deferment makes present frustration valuable by orienting players toward future competence. Communal Mythopoesis amplifies this core through shared ordeal stories, lore interpretation, and cultural memory. Through theoretical synthesis and comparative analysis of Dark Souls, Elden Ring, Hollow Knight, Lords of the Fallen, and The Surge, the paper shows how different configurations intensify, loosen, or undermine ordeal pleasure. OPF contributes a temporal account of competence frustration, a specific account of difficulty-coupled social meaning, and diagnostic concepts for designing difficult games that make frustration feel meaningful rather than arbitrary.
