Zurich Insurance Group (Switzerland)

As Fifth-generation (5G) networks advance towards future 6G technology, they provide enhanced capabilities in terms of data rates, latency, and connectivity. The expanding and heterogeneous landscape further extends the attack surface with increasing vulnerability to a wide range of dynamic cyber threats. Traditional Intrusion Detection Systems, relying on static signature-based methods, face significant challenges in detecting novel attacks in such complex environments. This paper proposes an explainable hybrid AE-XGBoost framework to address these gaps. In this proposed framework, the deep autoencoder serves as the backbone for two primary tasks: learning robust features from high-dimensional traffic data and synthesizing high-quality samples for minority, underrepresented attack classes. It effectively forms a class-balanced dataset. This is further extended to integrate Shapley Additive Explanations (SHAP), which provide needed interpretability. The framework robustness is determined by an extensive evaluation strategy. Extensive experiments are conducted using a 5-fold cross-validation and cross-dataset validation using three different datasets. The experimental results demonstrate the model effectiveness and promising generalization across the different datasets. Across these settings, the proposed IDS achieves F1-scores between approximately 0.96 and 0.999 and AUC values above 0.97, with only minor performance differences between original and synthetic test samples. SHAP-based analysis further reveals that a small subset of 5G- and TCP-level features consistently dominates the predictions, offering actionable explanations for security analysts. This provides a successful, scalable, and reliable security measure that proves effective and appropriate for 6G mobile network implementations in the near future.

Nonlinear model predictive control (MPC) and reinforcement learning (RL) are two powerful control strategies with complementary advantages. This work shows how actor-critic RL techniques can be leveraged to improve the performance of MPC. The RL critic is used as an approximation of the optimal value function, and an actor rollout provides an initial guess for the primal variables of the MPC. A parallel control architecture is proposed where each MPC instance is solved twice for different initial guesses. Besides the actor rollout initialization, a shifted initialization from the previous solution is used. The control actions from the lowest-cost trajectory are applied to the system at each time step. We provide some theoretical justification of the proposed algorithm by establishing that the discounted closed-loop cost is upper-bounded by the discounted closed-loop cost of the original RL actor plus an error term that depends on the (sub)optimality of the RL actor and the accuracy of the critic. These results do not require globally optimal solutions and indicate that larger horizons mitigate the effect of errors in the critic approximation. The proposed algorithm is intended for applications where standard methods to construct terminal costs or constraints for MPC are impractical. The approach is demonstrated in an illustrative toy example and an autonomous driving overtaking scenario.

BackgroundThe therapeutic potential of non-invasive transcranial electric stimulation (tES) is growing, but challenges in tailoring treatments to individual patients persist. We demonstrated that personalized montages, informed by biophysical modeling, can reduce dose variability and improve target engagement. However, traditional setups are prone to electrode placement errors, leading to inconsistent dosage, localization and potential off-target engagement.MethodsWe developed a pipeline for creating patient-specific 3D-printed caps based on individual MRI scans to allow custom placement of up to 34 electrodes. Hybrid brush electrodes were incorporated to facilitate the use on with-hair regions of the head. The patient-specific caps are integrated within a Neurostimulator/EEG device, enabling precise electrode placement for recording of neural activity and optimized, personalised tES therapies based on individual biophysical brain models.ResultsA proof-of-concept in-silico study highlighted the importance of positional accuracy in electrode placement, particularly in minimizing off-target stimulation in case of the personalized montage configuration. To mitigate positional displacement, we introduced the pipeline for generating patient-specific 3D-printed caps, ensuring precise anatomical fit. The flexible design of these caps supports high-density electrode arrays tailored to various therapeutic montages. Integrated brush electrodes function in both dry and wet modes and are powered by electronics embedded in a neckpiece, enabling multi-step tES and EEG recording with minimal setup for at-home use. Electrical and mechanical evaluations confirmed the system’s feasibility, and a clinical trial with eight healthy subjects showed improved comfort and increased willingness for home use.ConclusionThis customized cap-electrode system provides a practical solution for self-administered, repeatable personalized tES and EEG session at home.Trial registration: The study is registered on the Swiss National Clinical Trials Portal (SNCTP), (SNCTP000005598) and Clinicaltrials.gov (NCT05999916). The study was approved by the Independent Competent Ethics Committee (Ethikkommission Nordwest-und Zentralschweiz (EKNZ)) at the 14 August 2023 and the Competent Regulatory Authority Swissmedic at the 20 December 2023.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12984-025-01769-8.

Studies have implied that fibroblasts may act as regulators of immune cells in the tumor microenvironment (TME). We investigated the clinical relevance of fibroblast activation protein (FAP) positive stroma in high-grade serous ovarian cancer (HGSC) in relation to CD8+ lymphocyte's infiltration. In a discovery cohort (N=113) of HGSC, expression of FAP and CD8 in the TME was analyzed with immunohistochemistry. Results were correlated with overall survival (OS) and progression-free survival (PFS). The findings were validated in an independent cohort of HGSC (N=121) and in public available datasets. High infiltration of CD8+ cells in the TME of HGSC was found to be associated with longer OS, as previously known. Increased expression of FAP was associated with shorter median PFS (11.4 vs. 18.6months) in tumors with high density of CD8+ cells (HR 4.03, CI 95% 1.38-11.72, p=0.01). Similarly, in the validation cohort, high intensity of FAP in cases with high density of CD8+ cells was associated with shorter OS, 31.5 vs 76.9months (HR 2.83; 95% CI 1.17-6.86, p=0.02). The results were consistent in multivariable analyses. The association between high FAP expression and poor outcome in high density CD8 HGSC was also confirmed in publicly available datasets. The TME infiltration of FAP-positive fibroblasts is associated with poor prognosis in HGSC with high CD8+ cells density. Targeting the FAP+ subset of fibroblasts may unlock the local immune-activation in the TME thus enhance the positive prognostic effect of T-cells in ovarian cancer.

Abstract In order to design an Oxygen Reduction System (ORS) the most important parameter is the limiting oxygen concentration (LOC). Currently there are a couple of different small scale test methods used to determine the LOC for specific materials. In small scale tests an external heat flux is applied, to account for parameters such as geometry that otherwise cannot be replicated, theoretically arriving at a sufficiently low LOC. Other methods have tried to take heat transfer and geometry into consideration to better replicate real world conditions. However, there are some discussions into the validity to some of these existing methods. To replicate real world conditions a large scale setup was designed and used to determine LOC for certain warehouse configurations. This large scale setup is costly to run and therefore there is a need to investigate the possibility of creating an intermediate scale setup that has the ability to make flame propagation assessments for varying fuels, geometries and ignition sources. In this paper a first attempt for an intermediate scale setup is examined, and preliminary findings are discussed.

Abstract Marketplace lending has substantially changed since the first peer-to-peer lending platforms emerged in 2006. The industry is now an alternative to bank lending, predicted to total $70 billion for consumer and business loans worldwide by 2030. Marketplace lending is often deemed less safe than bank loans, mainly due to these portfolios' high degree of hidden information. These include needing more information on borrowers and potential correlations between them, which might lead to higher risk than is apparent at first glance. Deterministic processes cannot capture tail risk appropriately, so platforms and lenders should employ stochastic processes. This chapter introduces a Monte Carlo simulation and machine learning (ML) process to evaluate and monitor portfolios. For marketplace lending to become a viable and sustainable alternative to bank lending platforms, they must better evaluate, monitor, and manage tail risk in marketplace loans and develop tools to monitor and manage financial risk losses.

During the early stages of design, mechanisms are commonly modeled as perfect joints assembled with infinitely rigid bodies. This representation enables the prediction of the system’s mobilities through a mobility analysis. However, traditional mobility analysis tools can be computationally expensive or lack critical information, such as the type or direction of mobilities. It hinders the generation of topology and configuration through generative design schemes.In this paper, we propose an alternative approach to mobility analysis based on a real vector space representation of mobilities. Our method provides relevant information for early design steps while being computationally effective through a novel formulation of series and parallel assembly topological operations. A benchmark on four selected use cases highlights an acceleration of 3 to 4 orders of magnitude compared to traditional approaches. Additionally, design rules on the joints’ positions can be automatically generated with our approach. It enables the automation of the complete design process, including topology and configuration. As such, we provide guidelines to develop a generative design scheme dedicated to the synthesis of guiding mechanisms.

The capacity for terrestrial ecosystems to sequester additional carbon (C) with rising CO2 concentrations depends on soil nutrient availability1,2. Previous evidence suggested that mature forests growing on phosphorus (P)-deprived soils had limited capacity to sequester extra biomass under elevated CO2 (refs. 3–6), but uncertainty about ecosystem P cycling and its CO2 response represents a crucial bottleneck for mechanistic prediction of the land C sink under climate change7. Here, by compiling the first comprehensive P budget for a P-limited mature forest exposed to elevated CO2, we show a high likelihood that P captured by soil microorganisms constrains ecosystem P recycling and availability for plant uptake. Trees used P efficiently, but microbial pre-emption of mineralized soil P seemed to limit the capacity of trees for increased P uptake and assimilation under elevated CO2 and, therefore, their capacity to sequester extra C. Plant strategies to stimulate microbial P cycling and plant P uptake, such as increasing rhizosphere C release to soil, will probably be necessary for P-limited forests to increase C capture into new biomass. Our results identify the key mechanisms by which P availability limits CO2 fertilization of tree growth and will guide the development of Earth system models to predict future long-term C storage.