Norwegian Computing Center

The Cauchy-Schwarz (CS) divergence was developed by Príncipe et al. in 2000. In this paper, we extend the classic CS divergence to quantify the closeness between two conditional distributions and show that the developed conditional CS divergence can be elegantly estimated by a kernel density estimator from given samples. We illustrate the advantages (e.g., rigorous faithfulness guarantee, lower computational complexity, higher statistical power, and much more flexibility in a wide range of applications) of our conditional CS divergence over previous proposals, such as the conditional KL divergence and the conditional maximum mean discrepancy. We also demonstrate the compelling performance of conditional CS divergence in two machine learning tasks related to time series data and sequential inference, namely time series clustering and uncertainty-guided exploration for sequential decision making. The code of conditional CS divergence is available at https://github.com/SJYuCNEL/conditional_CS_divergence .

Improving probabilistic streamflow forecasts is critical for a multitude of water-oriented applications. Errors in water forecasts arise from several sources, one of which is the driving meteorology. Meteorological forecasts are often statistically post-processed before being input into hydrologic models. Shifts towards ensemble weather prediction systems have propelled advances in ensemble post-processing, providing an opportunity to enhance probabilistic water forecasting. This study’s purpose is to implement and evaluate the impact of coupling state-of-the-art precipitation ensemble post-processing techniques with the process-based, spatially distributed National Water Model (NWM). The post-processing has two steps: first, precipitation is calibrated using a censored, shifted, gamma distribution approach, and second, is reordered using an ensemble copula coupling technique. The NWM focuses on flood forecasting, but to date has only been run with time-lagged ensemble weather forecasts. We implement the NWM in a medium-range (∼7 day) ensemble forecasting mode for several rain-dominated catchments in northern California during an extremely wet water year, when advanced warning of heavy precipitation and streamflow could have been useful. Post-processing enhances NWM streamflow forecasts in terms of ensemble spread and accuracy, improving underestimation. Precipitation (streamflow) was generally skillful out to day 4 (7), including heavy precipitation (>75mm) and relatively high flow thresholds, but less consistently for the most extreme streamflow. These results suggest that NWM ensembles could be warranted for priority basins with relatively predictable weather phenomenon, though there are tradeoffs with hydrological model complexity and ensemble forecasting. This study can inform the NOAA-led Next Generation Water Resources Modeling Framework, which will need to consider how to integrate meteorological post-processing and ensemble techniques.

There has been much interest recently in implicit artificial intelligence (AI)-based approaches for geostatistical facies modeling. New generative machine learning constructions such as latent diffusion models (LDMs) appear to be competitive with traditional geostatistical approaches for facies characterization. Going beyond visual inspection of predictions, this study examines properties of the statistical distribution of samples generated by an LDM trained to generate facies models. The study uses a traditional truncated Gaussian random field (TGRF) model as a reference data-generating process and as the ground truth for benchmarking the LDM results. The distributions of realizations drawn from the LDM and TGRF models are compared using metrics including bias, variance, higher-order statistics, transiograms and Jensen–Shannon divergence for both marginal and joint (volume) distributions. Comparisons are made with and without conditioning on facies observations in wells for both stationary and nonstationary TGRF models with different covariance functions. The observed distributional differences are modest, and LDMs are regarded as a very promising approach here. Even so, some systematic artifacts are observed, such as underrepresentation of variability by the LDM. Moreover, the performance of the LDM is found to be sensitive to the training data.

Background Established assessment scales used for Parkinson’s disease (PD) have several limitations in tracking symptom progression and fluctuation. Both research and commercial-grade wearables show potential in improving these assessments. However, it is not known whether pervasive and affordable devices can deliver reliable data, suitable for designing open-source unobtrusive around-the-clock assessments. Our aim is to investigate the usefulness of the research-grade wristband Empatica E4, commercial-grade smartwatch Fitbit Sense, and the Oura ring, for PD research. Method The study included participants with PD (N = 15) and neurologically healthy controls (N = 16). Data were collected using established assessment scales (Movement Disorders Society Unified Parkinson’s Disease Rating Scale, Montreal Cognitive Assessment, REM Sleep Behavior Disorder Screening Questionnaire, Hoehn and Yahr Stage), self-reported diary (activities, symptoms, sleep, medication times), and 2-week digital data from the three devices collected simultaneously. The analyses comprised three steps: preparation (device characteristics assessment, data extraction and preprocessing), processing (data structuring and visualization, cross-correlation analysis, diary comparison, uptime calculation), and evaluation (usability, availability, statistical analyses). Results We found large variation in data characteristics and unsatisfactory cross-correlation. Due to output incongruences, only heart rate and movement could be assessed across devices. Empatica E4 and Fitbit Sense outperformed Oura in reflecting self-reported activities. Results show a weak output correlation and significant differences. The uptime was good, but Oura did not record heart rate and movement concomitantly. We also found variation in terms of access to raw data, sampling rate and level of device-native processing, ease of use, retrieval of data, and design. We graded the system usability of Fitbit Sense as good, Empatica E4 as poor, with Oura in the middle. Conclusions In this study we identified a set of characteristics necessary for PD research: ease of handling, cleaning, data retrieval, access to raw data, score calculation transparency, long battery life, sufficient storage, higher sampling frequencies, software and hardware reliability, transparency. The three analyzed devices are not interchangeable and, based on data features, none were deemed optimal for PD research, but they all have the potential to provide suitable specifications in future iterations.

Background Epidemiological studies link serum potassium (K⁺) to cognitive performance, but whether cognitive prognosis in dementia is related to K⁺ levels is unknown. Objective To determine if K⁺ levels predict cognitive prognosis in dementia and if this varies according to diagnosis or neuropathological findings. Methods This longitudinal cohort study recruited 183 patients with mild Alzheimer’s disease or Lewy body dementia (LBD). Serum K⁺ and eGFR were measured at baseline and medications which could affect K⁺ registered. The Mini-Mental State Examination (MMSE) was measured annually over 5 years, and mortality registered. Association between K⁺ and √(30 -MMSE) was estimated overall, and according to diagnosis (joint model). Associations between MMSE-decline and K⁺ were assessed in two subgroups with neuropathological examination (linear regression) or repeated measurements of K⁺ over 3 years (mixed model). Results Serum K⁺ at baseline was associated with more errors on MMSE over time (Estimate 0.18, p = 0.003), more so in LBD (p = 0.048). The overall association and LBD interaction were only significant in the 122 patients not using K⁺ relevant medication. Repeated K⁺ measures indicated that the association with MMSE errors over time was due to a between-person effect (p < 0.05, n = 57). The association between the annual MMSE decline was stronger in patients with autopsy confirmed LBD and more α-synuclein pathology (all: p < 0.05, n = 41). Conclusion Higher serum K⁺ predicts poorer cognitive prognosis in demented patients not using medications which affect K⁺, likely a between-person effect seen mainly in LBD.

“Just Accepted” papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To evaluate cancer detection and marker placement accuracy of two artificial intelligence (AI) models developed for interpretation of screening mammograms. Materials and Methods This retrospective study included data from 129 434 screening examinations (all female, mean age 59.2, SD = 5.8) performed between January 2008 and December 2018 in BreastScreen Norway. Model A was commercially available and B was an in-house model. Area under the receiver operating characteristic curve (AUC) with 95% confidence interval (CIs) were calculated. The study defined 3.2% and 11.1% of the examinations with the highest AI scores as positive, threshold 1 and 2, respectively. A radiologic review assessed location of AI markings and classified interval cancers as true or false negative. Results The AUC was 0.93 (95% CI: 0.92–0.94) for model A and B when including screen-detected and interval cancers. Model A identified 82.5% (611/741) of the screen-detected cancers at threshold 1 and 92.4% (685/741) at threshold 2. For model B, the numbers were 81.8% (606/741) and 93.7% (694/741), respectively. The AI markings were correctly localized for all screen-detected cancers identified by both models and 82% (56/68) of the interval cancers for model A and 79% (54/68) for B. At the review, 21.6% (45/208) of the interval cancers were identified at the preceding screening by either or both models, correctly localized and classified as false negative ( n = 17) or with minimal signs of malignancy ( n = 28). Conclusion Both AI models showed promising performance for cancer detection on screening mammograms. The AI markings corresponded well to the true cancer locations. ©RSNA, 2025

The protective effect of parity has been demonstrated for cancer of the breast, ovary, and endometrium but no studies have estimated the effect of each subsequent birth in women with 10 or more children or grand‐grand parity women, nor compared the linear relationship of the three cancers sites. Here, we aim to explore these relationships based on the Norwegian 1960 Census. The question of parity in present marriage was answered by 385,816 women born 1870–1915, a period with high fertility. Age at marriage has been validated as a proxy for age at first birth AFB. With high parity age at first birth will logically be restricted to early births giving structural zeros. Follow‐up was based on linkages to national registers until the first of any of the three diagnoses, death, or age 90 before 31.12.2005. Included were 16,905 breast cancers, 3827 ovarian cancers, and 3834 endometrial cancers. Age‐ and period‐specific incidence rates based on person‐years, PY, were used in logit regression models. The percentage decrease for each additional child over the total parity range was for breast cancer 10.5% (95% CI; 9.6–11.4), ovarian cancer 13.2% (11.2–15.3), and endometrial cancer 10.9% (8.9–12.8), in a model without higher order terms. Adjustment for structural zeros reduced the effect of age at first birth to less than one additional child. To the best of our knowledge this is the first analysis of the curvilinear relationships for cancers of the breast, ovary, and endometrium throughout the extended fertility range.

The expanding application of advanced analytics in insurance has generated numerous opportunities, such as more accurate predictive modeling powered by machine learning and artificial intelligence (AI) methods, the utilization of novel and unstructured datasets, and the automation of key operations. Significant advances in these areas are being made through novel applications and adaptations of predictive modeling techniques for insurance purposes, while, concurrently, rapid advances in machine learning methods are being made outside of the insurance sector. However, these innovations also bring substantial challenges, particularly around the transparency, explanation, and fairness of complex algorithmic models and the economic and societal impacts of their adoption in decision-making. As insurance is a highly regulated industry, models may be required by regulators to be explainable, in order to enable analysis of the basis for decision making. Due to the societal importance of insurance, significant attention is being paid to ensuring that insurance models do not discriminate unfairly. In this special issue, we feature papers that explore key issues in insurance analytics, focusing on prediction, explainability, and fairness.

Persons with autism spectrum disorder (ASD) and/or intellectual disability (ID) have difficulties in planning, organising and coping with change, which impedes the learning of daily living skills (DLSs), social participation and self-management across different environmental settings. Assistive technologies (ATs) is a broad term encompassing devices and services designed to support individuals with disabilities, and if used in a self-controlled manner, they may contribute inclusion in all domains of participation. This comprehensive literature review aims to critically assess and unify existing research that investigates the use of assistive technology within the practical domain for individuals with ASD and/or ID. The 18 relevant studies included in this review highlighted the benefits of AT for social participation and independence in daily activities of individuals with ASD and/or ID. Professionals working with this target group should be knowledgeable of the speedy progress of AT products and the potential of persons with ASD and/or ID to use mainstream devices to meet their individual needs. This awareness provides an opportunity to advocate for the universal benefits of AT for everyone. Technologies such as virtual reality, mobile applications and interactive software have been shown to improve DLSs, communication and social interaction. These tools offer engaging, user-friendly platforms that address the specific needs of these individuals, enhancing their learning and independence.

Associations for people with disabilities in Norway receive much feedback about negative experiences with travel. Little research has been done on this topic, and thus there is little knowledge about what can be done to improve these experiences. In this study, we have mapped travel experiences of people with disabilities and attitudes from employees in the tourism industry with two digital surveys. The questions were created in a workshop by collaboration with researchers, user representatives from a national association for people with disabilities, and employees from the tourism industry. The results show that some of the employees’ attitudes that are perceived as discriminatory by guests with disabilities are paradoxically caused by fear of doing something wrong. There seemed also to be a need for more knowledge about invisible disabilities, and a company-level and practice-based strategy for implementation of universal design in the customer service.

Artificial Intelligence (AI) holds significant potential for enhancing accessibility and user experience across digital products and services. However, mainstream web solutions commonly used by the general population still face accessibility barriers, hindering equal participation in the information society for people with disabilities. This article explores several promising applications of AI that can be used to create accessible solutions for people with disabilities. We also present our research, which aims to explore and demonstrate how various AI-based techniques can enhance and streamline accessibility testing for web solutions. We selected four success criteria from the Web Content Accessibility Guidelines (WCAG) that currently require extensive manual work and developed four prototypes using open-source machine-learning models to enhance conformance testing. While the prototypes need further optimization and evaluation, the results suggest that AI-based techniques can significantly reduce the need for manual work in accessibility testing.

Stochastic facies models based on truncated Gaussian random fields are known for being flexible and well suited to reproduce patterns and features from analogues or conceptual models. In pluri-Gaussian simulation, the number of random fields is theoretically unlimited, which adds flexibility and makes it possible to model a wider range of geological settings. However, the truncation map traditionally used to set up these models quickly becomes unclear when used for higher dimensions. Hence, in practical pluri-Gaussian applications, the number of fields is typically kept as low as two or three. We present a formulation of pluri-Gaussian simulation in which the truncation rule, the function that maps combinations of Gaussian random field values to facies categories, is represented as a particular binary tree. This is used to decouple the fields in the critical Gibbs sampling step of the conditioning process in such a way that we can use multiple lower-dimensional samples instead of a single higher-dimensional sample. The resulting conditioning algorithm scales excellently with the amount of conditioning data and the number of fields. The algorithm accepts a combination of trends and probabilities in the same model setup, which provides additional flexibility in representing varying depositional geometries. We demonstrate the hierarchical pluri-Gaussian simulation with two practical examples. One is based on real data from the Volve oil field in the North Sea. The other combines a large number of synthetic observations with a truncation tree tailored to a more complex geological concept. The choices made when building the truncation tree affect the features of the realizations, especially when it comes to which facies can be in contact and which can overprint each other. This aspect of tree building is discussed in light of the numerical examples given.

Postprocessing is a critical step in attaining calibrated and reliable probabilistic forecast output from numerical weather prediction models. A novel deep learning framework is proposed to postprocess 20 years of 7- and 14-day precipitation accumulation reforecasts from the Global Ensemble Forecast System at subseasonal time scales (week 1, week 2, and combined weeks 3–4 forecasts) over the contiguous United States. The network builds upon previous studies and is a combination of three parallel-trained components suitable for subseasonal prediction. The first is a ResUnet architecture which learns nonlinear relationships between binned observed precipitation and input images of weather and geographical variables. The second conditions the network to the month-of-year via a feature-wise linear modulation (FiLM) layer. The third helps the network learn when to revert the forecast to that of climatology. The RUFCO (named for its components ResUnet, FiLM, and Climatological-Offramp) forecasts are compared against raw and climatological forecasts as well as those from a state-of-the-art distributional regression postprocessing model, “censored, shifted gamma distribution (CSGD),” and a simple bias-corrected model. At week 1, every method exhibited a competitive advantage over climatological forecasts. At week 2, RUFCO generated forecasts with statistically significant improvement over climatology at 82%–94% of the domain, beating CSGD’s coverage of 76%–90% of the grid points. At week 3, RUFCO’s skillful coverage was 65%–85%, while CSGDs dropped to only 12%–37%. At the longer lead times, RUFCO achieved the highest domain-averaged skill scores across seasons. However, the network tends to “smooth” forecast skill, making it less competitive with CSGD in limited areas with strongly spatially varying biases. Significance Statement Precipitation accumulation forecasts 1, 2, and 3–4 weeks in advance are increasingly in-demand for a variety of decision-making applications around hydrologic forecasting, flood and drought awareness, and wildfire preparedness. However, raw forecasts from numerical weather prediction systems have errors that hinder skill. Postprocessing methods remove those errors and provide more reliable and skillful forecasts. We show that a new neural network technique is an effective and competitive postprocessing tool compared to more traditional techniques.

In our interconnected world, critical infrastructure systems are essential for modern societies. However, they face increasing threats, from cyberattacks to natural disasters. To address these challenges, the EU-CIP (Critical Infrastructure Protection) mission, a 3-year Coordination and Support Action funded by the European Commission, was initiated in October 2022. This chapter presents preliminary findings from the EU-CIP-ANALYSIS activities, shedding light on the evolving landscape of critical infrastructure protection. Preliminary findings reveal key needs, including adaptability, faster responses, and improved detection capabilities. Capability gaps include poor automation and inadequate control over interconnected assets. To address these needs and gaps, the EU-CIP mission has identified a range of state-of-the-art technologies and tools, including cyber-physical threat intelligence, security risk assessment, impact assessment tools, digital twins, anomaly detection, and machine learning for pattern detection. These technologies hold the potential to significantly enhance the resilience and security of critical infrastructure systems. This chapter provides valuable insights into the evolving landscape of critical infrastructure protection and resilience, highlighting the importance of innovative solutions and collaboration among stakeholders in safeguarding the vital systems that underpin modern societies.

Seismic data acquired at different times over the same area can provide insight into changes in an oil/gas reservoir. Probabilities for pore fluid will typically change while the lithology remains stable over time. This implies significant correlations across the vintages. We have developed a methodology for Bayesian prediction of joint probabilities for discrete lithology-fluid classes (LFCs) for two vintages, simultaneously considering the seismic AVO data of both vintages. By taking into account cross-vintage correlations of elastic and seismic properties, the simultaneous invertion ensures that the individual results for both vintages as well as their differences are consistent and constrained by the seismic data of both vintages. The method relies on prior geological knowledge of stratigraphic layering, possible lithologies and fluids within each layer, and possible cross-vintage changes in lithology and pore fluid. Multiple LFCs can be used to represent different strengths of dynamic cross-vintage changes. We have tested the algorithm on a synthetic data set and on data from the Edvard Grieg field in the central North Sea. Synthetic results demonstrate that the algorithm is able to use dual-vintage data together with a prior model specifying their correlations to calculate joint LFC posterior probabilities for both vintages with a lower degree of uncertainty than independent single-vintage inversions. The Edvard Grieg results indicate that the underlying model is sufficiently general to explain 4D variations in seismic data using a reasonably simple prior model of 4D LFC changes.

The availability of textual data depicting human-centered features and behaviors is crucial for many data mining and machine learning tasks. However, data containing personal information should be anonymized prior making them available for secondary use. A variety of text anonymization methods have been proposed in the last years, which are standardly evaluated by comparing their outputs with human-based anonymizations. The residual disclosure risk is estimated with the recall metric, which quantifies the proportion of manually annotated re-identifying terms successfully detected by the anonymization algorithm. Nevertheless, recall is not a risk metric, which leads to several drawbacks. First, it requires a unique ground truth, and this does not hold for text anonymization, where several masking choices could be equally valid to prevent re-identification. Second, it relies on human judgements, which are inherently subjective and prone to errors. Finally, the recall metric weights terms uniformly, thereby ignoring the fact that the influence on the disclosure risk of some missed terms may be much larger than of others. To overcome these drawbacks, in this paper we propose a novel method to evaluate the disclosure risk of anonymized texts by means of an automated re-identification attack. We formalize the attack as a multi-class classification task and leverage state-of-the-art neural language models to aggregate the data sources that attackers may use to build the classifier. We illustrate the effectiveness of our method by assessing the disclosure risk of several methods for text anonymization under different attack configurations. Empirical results show substantial privacy risks for most existing anonymization methods.

During the COVID-19 pandemic in Norway, the testing criteria and capacity changed numerous times. In this study, we aim to assess consequences of changes in testing criteria for infectious disease surveillance. We plotted the proportion of positive PCR tests and the total number of PCR tests for different periods of the pandemic in Norway. We fitted regression models for the total number of PCR tests and the probability of positive PCR tests, with time and weekday as explanatory variables. The regression analysis focuses on the time period until 2021, i.e. before Norway started vaccination. There were clear changes in testing criteria and capacity over time. In particular, there was a marked difference in the testing regime before and after the introduction of self-testing, with a drastic increase in the proportion of positive PCR tests after the introduction of self-tests. The probability of a PCR test being positive was higher for weekends and public holidays than for Mondays-Fridays. The probability for a positive PCR test was lowest on Mondays. This implies that there were different testing criteria and/or different test-seeking behaviour on different weekdays. Though the probability of testing positive clearly changed over time, we cannot in general conclude that this occurred as a direct consequence of changes in testing policies. It is natural for the testing criteria to change during a pandemic. Though smaller changes in testing criteria do not seem to have large, abrupt consequences for the disease surveillance, larger changes like the introduction and massive use of self-tests makes the test data less useful for surveillance.

Aerial drone imaging is an efficient tool for mapping and monitoring of coastal habitats at high spatial and temporal resolution. Specifically, drone imaging allows for time- and cost-efficient mapping covering larger areas than traditional mapping and monitoring techniques, while also providing more detailed information than those from airplanes and satellites, enabling for example to differentiate various types of coastal vegetation. Here, we present a systematic method for shallow water habitat classification based on drone imagery. The method includes:•Collection of drone images and creation of orthomosaics. •Gathering ground-truth data in the field to guide the image annotation and to validate the final map product. •Annotation of drone images into – potentially hierarchical – habitat classes and training of machine learning algorithms for habitat classification. As a case study, we present a field campaign that employed these methods to map a coastal site dominated by seagrass, seaweed and kelp, in addition to sediments and rock. Such detailed but efficient mapping and classification can aid to understand and sustainably manage ecologically and valuable marine ecosystems.