L3S Research Center

Multiple system atrophy (MSA) is a neurodegenerative disorder characterized by 3 core symptom complexes: parkinsonism, cerebellar syndrome, and dysautonomia. Recent Movement Disorder Society (MDS) criteria allow for the clinical diagnosis of MSA based solely on motor symptoms, without requiring dysautonomia. This study aimed to evaluate the frequency and disease trajectory of MSA patients without dysautonomia compared with those with autonomic involvement. A multicenter cohort of autopsy-confirmed patients with MSA was analyzed for demographic characteristics, symptom onset, and progression of parkinsonism, cerebellar syndrome, and dysautonomia. Clinical data were collected through standardized chart reviews across participating centers and categorized using the MDS-MSA criteria. Patients were grouped according to their initial symptom complex and tracked for the evolution of additional symptoms. Analyses included time to development of further symptom complexes, age at symptom onset, disease duration, and phenotype at the last recorded visit. Patients with motor symptoms only were matched to patients with similar demographics but with dysautonomia. Statistical methods included ANOVA, t tests, Welch t tests, and χ2 tests, with significance set at p < 0.05. Among 140 patients (mean age at onset 62.3 ± 8.9 years; 44% female), 81 (58%) initially presented without dysautonomia (57 with parkinsonism only, 17 with cerebellar syndrome only, 7 with both). At final follow-up, 12 patients (9%) had not developed dysautonomia. These patients showed significantly longer disease duration (mean 8.1 ± 2.1 years) than matched patients with dysautonomia (mean 6.3 ± 2.6 years; p = 0.035). Overall, 51% of patients developed all 3 symptom complexes. Patients with cerebellar onset progressed more rapidly to multisystem involvement than those with parkinsonian onset (mean interval to second symptom: 2.0 vs 3.4 years; p < 0.05). The MDS-MSA criteria expand the diagnostic scope by identifying a motor-only subgroup with a distinct and potentially slower disease course. These findings underscore the importance of including motor-only patients in natural history and interventional studies. Limitations include retrospective data collection and potential variability in symptom documentation.

Across essentially all fields of research, many aspects of the respective research processes – whether experimental, theoretical, empirical or outright computational – are closely related to software. Yet the process of finding software that is directly suitable or at least a good starting point for a given research task is cumbersome. This project aims to develop a community-driven system that provides potential users of research software with a diversity of pathways towards actually finding software that closely matches their research needs if such software exists. Conversely, it will provide software developers with mechanisms to make their software findable for research-related tasks and it will highlight mismatches between software supply and demand for specific tasks. To this end, we will document how various stakeholders of the research landscape have been searching for – or stumbling upon – research software so far, identify variables associated with successful search outcomes and build workflows that assist in describing software and associated concepts in a standardised fashion. These descriptions will then be aligned across various sources of relevant information and integrated into Wikidata, the knowledge graph that anyone can edit and that already contains considerable breadth and depth of information related to research, software and their interactions. While keeping an eye on similar approaches to software discovery that might work in parts of the research ecosystem, existing Wikidata content and workflows will be reviewed and built upon. Additional documentation, tooling and workflows will be developed to enrich, expand, curate, query and explore this content, both for specific use cases and with ongoing engagement of the communities involved in research software, open data or collaborative curation. Within its three years, the project seeks to establish a dedicated community overseeing a well-documented and smoothly running infrastructure for software discovery and to devise a plan for how this can be sustained for the longer term.

Abstract Increasing competition in science and higher education has sparked contentious debate and scholarly attention in recent years. Despite their significant merits, prevailing accounts of competition fall short of capturing the competitive dynamics in science and higher education: Most of the time, actors in this sector are involved in not just one but several competitions. Universities as organizations, researchers as individual actors, and also state actors are simultaneously embedded in different, nested, and interdependent competitions, which we refer to as multiple competition. Individual and collective actors engage in heterogeneous—albeit interrelated—forms of competition for scarce symbolic and material goods like attention, reputation, ranking positions, research grants, high-quality publications, personnel, and employment. Furthermore, the multiplicity of competitions that individual academics, universities, and state agencies face might reinforce each other. Using theoretical resources from sociology and economics, we propose a new conceptual framework for analyzing constellations of multiple competition in science and higher education. We demonstrate the added value of this conceptualization for empirical studies by drawing on examples from different academic systems.

Cyber-physical systems (CPSs) must meet safety-critical requirements and optimization goals. Reactive controller synthesis (RCS) from temporal logic (TL) specifications can automatically construct agent behaviors that ensure safety and liveness. However, RCS struggles with time and probability aspects. Reinforcement Learning (RL) can optimize system behaviors, but cannot guarantee correctness. RL and RCS can be combined by shielding an agent with a synthesized controller, ensuring that the agent behavior satisfies the TL specification. However, this requires aligning different models: the TL specification and the often hand-crafted RL training environment code. We propose a three-step method, centered around code generation and manual refinement, for streamlining this alignment: (1) Formalize discrete environment assumptions and guarantees into a Spectra Tl specification. (2) From it, automatically generate Gymnasium-compatible training environment code. For the target code, we employ BPPY which enables a unique one-to-one mapping of the Spectra properties to individual code modules. This allows users to refine time and probability aspects as well as optimization goals without misaligning the models. (3) Synthesize a permissive controller from the TL specification and integrate it as a shield for the agent. In contrast to previous work, our approach (a) guarantees not only safety but also liveness; (b) even if the system is too complex for RCS, we can leverage the generated BPPY code as a model-at-runtime that shields the agent from unsafe actions.

Urban Foundation Models (UFMs) are emerging as a transformative paradigm for understanding and managing complex urban systems, offering new capabilities to improve and transform urban environments. Trajectory data (e.g., vehicle movements) contains rich spatio-temporal patterns of human mobility and urban dynamics that are crucial for understanding city-wide movement behaviors, traffic patterns, and transportation system operations. Effectively modeling such trajectory data is crucial for UFMs, yet this poses significant challenges due to the high-dimensional nature of raw trajectories and their complex spatio-temporal dynamics. Trajectory representation learning addresses these challenges by transforming raw trajectory data into compact, meaningful representations that capture essential mobility patterns and facilitates the application to various urban tasks, such as trajectory similarity computation, travel time estimation, or destination prediction. This is achieved by learning low-dimensional representations from high-dimensional and raw trajectory data. However, existing methods for trajectory representation learning either rely on grid-based or road-based representations, which are inherently different and thus, could lose information contained in the other modality. Moreover, these methods overlook the dynamic nature of urban traffic, relying on static road network features rather than time-varying traffic patterns. In this paper, we propose TIGR , a novel model designed to integrate grid and road network modalities while incorporating spatio-temporal dynamics to learn rich, general-purpose representations of trajectories. We evaluate TIGR on three real-world datasets and demonstrate the effectiveness of combining both modalities by substantially outperforming state-of-the-art methods, i.e., up to 59.34% for trajectory similarity, up to 16.65% for travel time estimation, and up to 14.81% for destination prediction.

Abstract. The dynamic response of wind turbine rotor blades can result in multi-axial, non-proportional stress histories in the adhesive joints. These are not properly considered in current design guidelines and standards. In this paper, the concept of non-proportionality factors is utilized for the quantification of the degree of non-proportionality. A novel formulation is proposed and verified, followed by its application to the analysis of trailing edge adhesive joints of three different blade designs with varying blade sizes. Two simple fatigue analysis concepts, one based on global equivalent stresses and the other on the search for a critical plane, are evaluated with respect to their ability to account for multi-axial, non-proportional stress time series. A clear correlation of the degree of non-proportionality with the blade size was not found, as the blade response depends on not only the turbine size but also the blade design philosophy. A remarkable degree of non-proportionality was observed for all three blades. The fatigue damage criteria employed in this study revealed that the critical-plane approach can deal with non-proportional, multi-axial stress time series and therefore is more precise when non-proportionality is involved. The global equivalent stress approach seems to provide conservative predictions. However, there are implications from the literature that global equivalent stress approaches can result in non-conservative estimates under specific circumstances, so further research including validation of appropriate models is necessary for final conclusions.

The Second Search Futures Workshop, in conjunction with the Forty-seventh European Conference on Information Retrieval (ECIR) 2025, looked into the future of search to ask questions such as: • How can we navigate data privacy in large language model (LLM)-based information retrieval (IR)? • How can we implement agentic IR for proactive knowledge synthesis? • How do we ensure trustworthy information access beyond citations in the age of language models? • How does deep search transition from matching to reasoning? • What is meant by information semantics, knowledge representation, and natural language in a world of LLM-powered search? • What are serendipity engines, and how do they explore proactive web search via LLM agents, retrieval augmented generation (RAG), and simulated user feedback? The second edition of the workshop opened with ten lightning talks from a diverse group of speakers. Rather than traditional paper presentations, these short talks offered concise overviews of emerging ideas and critical insights, enabling a rapid exchange across various topics. The format was designed to spark discussion and expose participants to a broad spectrum of future-facing research directions in a compact timeframe. This report, co-authored by the workshop organizers, presenters, and participants, summarizes the talks and key discussions. Our aim is to share these insights with the broader IR community and help seed further dialogue around the themes raised. Date: 10 April 2025. Website: https://searchfutures.github.io/.