Search

Neural networks have shown remarkable performance in computer vision, but their deployment in real-world scenarios is challenging due to their sensitivity to image corruptions. Existing attribution methods are uninformative for explaining the sensitivity to image corruptions, while the literature on robustness only provides model-based explanations. However, the ability to scrutinize models' behavior under image corruptions is crucial to increase the user's trust. Towards this end, we introduce the Wavelet sCale Attribution Method (WCAM), a generalization of attribution from the pixel domain to the space-scale domain. Attribution in the space-scale domain reveals where and on what scales the model focuses. We show that the WCAM explains models' failures under image corruptions, identifies sufficient information for prediction, and explains how zoom-in increases accuracy., Comment: main: 9 pages, appendix 19 pages, 32 figures, 5 tables

Live biotherapeutic products constitute an emerging therapeutic approach to prevent or treat inflammatory bowel diseases. Lactobacillus acidophilus is a constituent of the human microbiota with probiotic potential, that is illustrated by improvement of intestinal inflammation and antimicrobial activity against several pathogens. In this study, we evaluated the immunomodulatory properties of the L. acidophilus strain BIO5768 at steady state and upon acute inflammation. Supplementation of naïve mice with BIO5768 heightened the transcript level of some IL-17 target genes encoding for protein with microbicidal activity independently of NOD2 signaling. Of these, the BIO5768-induced expression of Angiogenin-4 was blunted in monocolonized mice that are deficient for the receptor of IL-17 (but not for NOD2). Interestingly, priming of bone marrow derived dendritic cells by BIO5768 enhanced their ability to support the secretion of IL-17 by CD4+ T cells. Equally of importance, the production of IL-22 by type 3 innate lymphoid cells is concomitantly heightened in response to BIO5768. When administered alone or in combination with Bifidobacterium animalis spp. lactis BIO5764 and Limosilactobacillus reuteri, BIO5768 was able to alleviate at least partially intestinal inflammation induced by Citrobacter rodentium infection. Furthermore, BIO5768 was also able to improve colitis induced by 2,4,6-trinitrobenzene sulfonic acid (TNBS). In conclusion, we identify a new potential probiotic strain for the management of inflammatory bowel diseases, and provide some insights into its IL-17-dependent and independent mode of action.

Photovoltaic (PV) energy is rapidly growing and key to mitigating the energy crisis. However, distributed PV generation, which amounts to half of the PV installed capacity, is typically unavailable to transmission system operators (TSOs), making it increasingly difficult to balance the load and supply and avoid grid congestions. To assess distributed PV generation, TSOs need precise knowledge regarding the metadata of distributed PV installations. Many remote sensing-based approaches have been proposed to map these installations in recent years. However, to use these methods in industrial processes, assessing their accuracy over the mapping area, i.e., the area covered by the model during deployment, is necessary. We define the downstream task accuracy (DTA) as the accuracy over the mapping area, automatically computed using publicly available data sources and the model's outputs and expressed in an interpretable way for operators. We benchmark existing models for distributed PV mapping and show how they perform in terms of DTA. We show that the accuracy computed on the test set overestimates by about 30 percentage points the accuracy on the mapping area. Our approach paves the way for safer integration of deep-learning-based pipelines for remote PV mapping. Code is available at https://github.com/gabrielkasmi/deeppvmapper., Comment: 12 pages, 2 figures, 3 tables. Camera-ready version, accepted for Workshop on Machine Learning for Earth Observation (MACLEAN), in Conjunction with the ECML/PKDD 2022. Final paper version, as published on CEUR-WS http://ceur-ws.org/Vol-3343/

Lavrik and the author gave uniform bounds of the error term in the approximate functional equation for the derivatives of the Hardy's Z-function. We obtain a new bound of this error term which is much better for high order derivatives., Comment: 10 pages

This contribution reports the archaeometric study of 62 fragmented sculptures from the collections stored in the reserves of Villa Adriana, Tivoli. They were carved in white fine to very fine grained marble, but also medium and coarse types were investigated. Provenance determination was carried out using a sequential approach, taking into account petrography and cathodomicrofacies as the first step, combined with the C and O stable isotopes. Complementary techniques, EPR and quantitative CL,were used to verify or specify from among the different options. The paper discusses the results of combining different analytical techniques in comparison to the most reliable databases. They confirm the use of Carrara but also point to a variety of classical marble sources in Greece and Turkey, including the recently discovered Göktepe quarries. Some results must be considered provisional, until the detailed petrography of Göktepe is published.

The aim of this research is to establish a relation between the derivatives of Hardy's Z function and the argument of the Riemann zeta function in the neighborhood of points where |Z| reaches a large maximum. In this paper, we make a step toward this goal by solving a problem of the same nature.

Dissertation, RWTH Aachen University, 2019; Aachen 1 Online-Ressource (187 Seiten) : Illustrationen (2019). = Dissertation, RWTH Aachen University, 2019, The cloud induced variability of the solar resource is an operational challenge for both photovoltaic and concentrating solar plants. This and their growing contributions to overall power generation challenge the stability of electricity grids. Storage and grid development can be used to address this issue, whose deployments are optimized by forecasts. For several days ahead, solar forecasts are provided by Numerical Weather Prediction models with typical resolutions of 16-50 km and hours to days. Weather satellites provide solar irradiance predictions with kilometric spatial resolutions and typical temporal resolutions between 5 min and 15 min. This leaves a gap for local, short-term and high resolution solar forecasting systems. High resolution forecasts for the next minutes ahead are notably provided by all-sky imager (ASI) based nowcasting systems. Such systems consist of at least one camera taking hemispherical images of the sky. A large amount of hardware and software configurations to derive such short-term forecasts is proposed in literature. However, different validation periods, locations and approaches as well as the lack of comprehensive benchmarks make direct comparisons difficult. Therefore, the optimal configuration of ASI based nowcasting systems is unclear. To address this question, a modular framework to achieve high resolution spatially resolved solar forecasts for irradiance maps is developed. Within this framework, different hardware and software configurations are tested, highlighting the need for comprehensive benchmarks for each main step involved in the processing: cloud segmentation, cloud motion vector assessment, cloud height estimation. Four cloud segmentation approaches are benchmarked. Furthermore, a novel tool to validate cloud motion vector measurements is developed and applied, ruling out certain setups. Comparing five different cloud height measurement approaches found a setup based on two ASIs to be the most promising. The optimal distance between the cameras of such a two ASI system is investigated both with an in-field study and modeling. Besides benchmarking individual sub-tasks of nowcasting systems, a framework to evaluate nowcasted irradiance maps is applied to example configurations. Special focus is put on temporal and spatial aggregation effects. In order to investigate spatial aggregation effects, spatially resolved reference irradiance maps provided by a developed shadow camera system are compared to nowcasted irradiance maps. Aggregation effects are found to have a strong effect on deviations. With the implemented frameworks and conducted benchmarks, the optimal system among the considered configurations to nowcast irradiance maps is determined to be a two camera setup using a clear sky library based cloud segmentation as well as novel differential approaches for cloud height and motion vector estimations. Further options for improvements are identified to include in-depth studies of cloud dynamics, combinations of ASIs with other sensors such as satellites, optimized cloud segmentation algorithms and the development of systems dedicated to solar ramp forecasts., Published by Aachen