@article{GairingHoegeleKosenkova2017, author = {Gairing, Jan and H{\"o}gele, Michael and Kosenkova, Tetiana}, title = {Transportation distances and noise sensitivity of multiplicative Levy SDE with applications}, series = {Stochastic processes and their application}, volume = {128}, journal = {Stochastic processes and their application}, number = {7}, publisher = {Elsevier}, address = {Amsterdam}, issn = {0304-4149}, doi = {10.1016/j.spa.2017.09.003}, pages = {2153 -- 2178}, year = {2017}, abstract = {This article assesses the distance between the laws of stochastic differential equations with multiplicative Levy noise on path space in terms of their characteristics. The notion of transportation distance on the set of Levy kernels introduced by Kosenkova and Kulik yields a natural and statistically tractable upper bound on the noise sensitivity. This extends recent results for the additive case in terms of coupling distances to the multiplicative case. The strength of this notion is shown in a statistical implementation for simulations and the example of a benchmark time series in paleoclimate.}, language = {en} } @article{BuergerPfisterBronstert2019, author = {B{\"u}rger, Gerd and Pfister, A. and Bronstert, Axel}, title = {Temperature-Driven Rise in Extreme Sub-Hourly Rainfall}, series = {Journal of climate}, volume = {32}, journal = {Journal of climate}, number = {22}, publisher = {American Meteorological Soc.}, address = {Boston}, issn = {0894-8755}, doi = {10.1175/JCLI-D-19-0136.1}, pages = {7597 -- 7609}, year = {2019}, abstract = {Estimates of present and future extreme sub-hourly rainfall are derived from a daily spatial followed by a sub-daily temporal downscaling, the latter of which incorporates a novel, and crucial, temperature sensitivity. Specifically, daily global climate fields are spatially downscaled to local temperature T and precipitation P, which are then disaggregated to a temporal resolution of 10 min using a multiplicative random cascade model. The scheme is calibrated and validated with a group of 21 station records of 10-min resolution in Germany. The cascade model is used in the classical (denoted as MC) and in the new T-sensitive (MC+) version, which respects local Clausius-Clapeyron (CC) effects such as CC scaling. Extreme P is positively biased in both MC versions. Observed T sensitivity is absent in MC but well reproduced by MC+. Long-term positive trends in extreme sub-hourly P are generally more pronounced and more significant in MC+ than in MC. In units of 10-min rainfall, observed centennial trends in annual exceedance counts (EC) of P > 5 mm are +29\% and in 3-yr return levels (RL) +27\%. For the RCP4.5-simulated future, higher extremes are projected in both versions MC and MC+: per century, EC increases by 30\% for MC and by 83\% for MC+; the RL rises by 14\% for MC and by 33\% for MC+. Because the projected daily P trends are negligible, the sub-daily signal is mainly driven by local temperature.}, language = {en} } @article{ThonLandwehrDeRaedt2011, author = {Thon, Ingo and Landwehr, Niels and De Raedt, Luc}, title = {Stochastic relational processes efficient inference and applications}, series = {Machine learning}, volume = {82}, journal = {Machine learning}, number = {2}, publisher = {Springer}, address = {Dordrecht}, issn = {0885-6125}, doi = {10.1007/s10994-010-5213-8}, pages = {239 -- 272}, year = {2011}, abstract = {One of the goals of artificial intelligence is to develop agents that learn and act in complex environments. Realistic environments typically feature a variable number of objects, relations amongst them, and non-deterministic transition behavior. While standard probabilistic sequence models provide efficient inference and learning techniques for sequential data, they typically cannot fully capture the relational complexity. On the other hand, statistical relational learning techniques are often too inefficient to cope with complex sequential data. In this paper, we introduce a simple model that occupies an intermediate position in this expressiveness/efficiency trade-off. It is based on CP-logic (Causal Probabilistic Logic), an expressive probabilistic logic for modeling causality. However, by specializing CP-logic to represent a probability distribution over sequences of relational state descriptions and employing a Markov assumption, inference and learning become more tractable and effective. Specifically, we show how to solve part of the inference and learning problems directly at the first-order level, while transforming the remaining part into the problem of computing all satisfying assignments for a Boolean formula in a binary decision diagram. We experimentally validate that the resulting technique is able to handle probabilistic relational domains with a substantial number of objects and relations.}, language = {en} } @article{SysoevPonomarenkoPikovskij2017, author = {Sysoev, Ilya V. and Ponomarenko, Vladimir I. and Pikovskij, Arkadij}, title = {Reconstruction of coupling architecture of neural field networks from vector time series}, series = {Communications in nonlinear science \& numerical simulation}, volume = {57}, journal = {Communications in nonlinear science \& numerical simulation}, publisher = {Elsevier}, address = {Amsterdam}, issn = {1007-5704}, doi = {10.1016/j.cnsns.2017.10.006}, pages = {342 -- 351}, year = {2017}, abstract = {We propose a method of reconstruction of the network coupling matrix for a basic voltage-model of the neural field dynamics. Assuming that the multivariate time series of observations from all nodes are available, we describe a technique to find coupling constants which is unbiased in the limit of long observations. Furthermore, the method is generalized for reconstruction of networks with time-delayed coupling, including the reconstruction of unknown time delays. The approach is compared with other recently proposed techniques.}, language = {en} } @article{SchwiederWesemeyerFrantzetal.2022, author = {Schwieder, Marcel and Wesemeyer, Maximilian and Frantz, David and Pfoch, Kira and Erasmi, Stefan and Pickert, J{\"u}rgen and Nendel, Claas and Hostert, Patrick}, title = {Mapping grassland mowing events across Germany based on combined Sentinel-2 and Landsat 8 time series}, series = {Remote sensing of environment}, volume = {269}, journal = {Remote sensing of environment}, publisher = {Elsevier}, address = {New York}, issn = {0034-4257}, doi = {10.1016/j.rse.2021.112795}, pages = {16}, year = {2022}, abstract = {Spatially explicit knowledge on grassland extent and management is critical to understand and monitor the impact of grassland use intensity on ecosystem services and biodiversity. While regional studies allow detailed insights into land use and ecosystem service interactions, information on a national scale can aid biodiversity assessments. However, for most European countries this information is not yet widely available. We used an analysis-ready-data cube that contains dense time series of co-registered Sentinel-2 and Landsat 8 data, covering the extent of Germany. We propose an algorithm that detects mowing events in the time series based on residuals from an assumed undisturbed phenology, as an indicator of grassland use intensity. A self-adaptive ruleset enabled to account for regional variations in land surface phenology and non-stationary time series on a pixelbasis. We mapped mowing events for the years from 2017 to 2020 for permanent grassland areas in Germany. The results were validated on a pixel level in four of the main natural regions in Germany based on reported mowing events for a total of 92 (2018) and 78 (2019) grassland parcels. Results for 2020 were evaluated with combined time series of Landsat, Sentinel-2 and PlanetScope data. The mean absolute percentage error between detected and reported mowing events was on average 40\% (2018), 36\% (2019) and 35\% (2020). Mowing events were on average detected 11 days (2018), 7 days (2019) and 6 days (2020) after the reported mowing. Performance measures varied between the different regions of Germany, and lower accuracies were found in areas that are revisited less frequently by Sentinel-2. Thus, we assessed the influence of data availability and found that the detection of mowing events was less influenced by data availability when at least 16 cloud-free observations were available in the grassland season. Still, the distribution of available observations throughout the season appeared to be critical. On a national scale our results revealed overall higher shares of less intensively mown grasslands and smaller shares of highly intensively managed grasslands. Hotspots of the latter were identified in the alpine foreland in Southern Germany as well as in the lowlands in the Northwest of Germany. While these patterns were stable throughout the years, the results revealed a tendency to lower management intensity in the extremely dry year 2018. Our results emphasize the ability of the approach to map the intensity of grassland management throughout large areas despite variations in data availability and environmental conditions.}, language = {en} } @article{PohleAdamBeumer2022, author = {Pohle, Jennifer and Adam, Timo and Beumer, Larissa}, title = {Flexible estimation of the state dwell-time distribution in hidden semi-Markov models}, series = {Computational statistics \& data analysis}, volume = {172}, journal = {Computational statistics \& data analysis}, publisher = {Elsevier}, address = {Amsterdam}, issn = {0167-9473}, doi = {10.1016/j.csda.2022.107479}, pages = {15}, year = {2022}, abstract = {Hidden semi-Markov models generalise hidden Markov models by explicitly modelling the time spent in a given state, the so-called dwell time, using some distribution defined on the natural numbers. While the (shifted) Poisson and negative binomial distribution provide natural choices for such distributions, in practice, parametric distributions can lack the flexibility to adequately model the dwell times. To overcome this problem, a penalised maximum likelihood approach is proposed that allows for a flexible and data-driven estimation of the dwell-time distributions without the need to make any distributional assumption. This approach is suitable for direct modelling purposes or as an exploratory tool to investigate the latent state dynamics. The feasibility and potential of the suggested approach is illustrated in a simulation study and by modelling muskox movements in northeast Greenland using GPS tracking data. The proposed method is implemented in the R-package PHSMM which is available on CRAN.}, language = {en} } @article{SchneiderWenigPapenbrock2021, author = {Schneider, Johannes and Wenig, Phillip and Papenbrock, Thorsten}, title = {Distributed detection of sequential anomalies in univariate time series}, series = {The VLDB journal : the international journal on very large data bases}, volume = {30}, journal = {The VLDB journal : the international journal on very large data bases}, number = {4}, publisher = {Springer}, address = {Berlin}, issn = {1066-8888}, doi = {10.1007/s00778-021-00657-6}, pages = {579 -- 602}, year = {2021}, abstract = {The automated detection of sequential anomalies in time series is an essential task for many applications, such as the monitoring of technical systems, fraud detection in high-frequency trading, or the early detection of disease symptoms. All these applications require the detection to find all sequential anomalies possibly fast on potentially very large time series. In other words, the detection needs to be effective, efficient and scalable w.r.t. the input size. Series2Graph is an effective solution based on graph embeddings that are robust against re-occurring anomalies and can discover sequential anomalies of arbitrary length and works without training data. Yet, Series2Graph is no t scalable due to its single-threaded approach; it cannot, in particular, process arbitrarily large sequences due to the memory constraints of a single machine. In this paper, we propose our distributed anomaly detection system, short DADS, which is an efficient and scalable adaptation of Series2Graph. Based on the actor programming model, DADS distributes the input time sequence, intermediate state and the computation to all processors of a cluster in a way that minimizes communication costs and synchronization barriers. Our evaluation shows that DADS is orders of magnitude faster than S2G, scales almost linearly with the number of processors in the cluster and can process much larger input sequences due to its scale-out property.}, language = {en} }