Dr. Folke Mitzlaff

Author summary Many species occasionally hybridise and share genetic material with related species. Interspecific gene flow may be counteracted by natural selection at particular ‘barrier loci’. As a result, a pair of species can end up sharing more genetic variation in some parts of their genome than in others, and the tree of relationships in a group of species can differ from one part of the genome to another. We studied relationships and barriers among three species of Heliconius butterflies using whole-genome sequences from nine populations. We find that species relationships vary dramatically and predictably across the genome because the species barriers are more porous in genomic regions with higher recombination rates. This occurs because recombination determines how broadly the surrounding genome is affected by a barrier locus. The genome-wide pattern suggests that barrier loci are widespread across the genome. One consequence is that smaller chromosomes, which have higher recombination rates, tend to have weaker species barriers than longer chromosomes. The relationships among populations on small chromosomes therefore tend to be predicted by geography, rather than by which species they belong to. Our work shows how hybridisation, recombination, and selection interact to reshape species’ relationships.

This study presents a novel application of machine learning to deliver optimised, multi-model combinations (MMCs) of Global Hydrological Model (GHM) simulations. We exemplify the approach using runoff simulations from five GHMs across 40 large global catchments. The benchmarked, median performance gain of the MMC solutions is 45% compared to the best performing GHM and exceeds 100% when compared to the ensemble mean (EM). The performance gain offered by MMC suggests that future multi-model applications consider reporting MMCs, alongside the EM and intermodal range, to provide end-users of GHM ensembles with a better contextualised estimate of runoff. Importantly, the study highlights the difficulty of interpreting complex, non-linear MMC solutions in physical terms. This indicates that a pragmatic approach to future MMC studies based on machine learning methods is required, in which the allowable solution complexity is carefully constrained.

We derive Bayesian distances for all stars in the RV sample of Gaia DR2, and use the statistical method of Schoenrich, Binney & Asplund(2012) to validate the distances and test the Gaia parallaxes. In contrast to other methods, which rely on special sources, our method directly tests the distances to all stars in our sample. We find clear evidence for a near-linear trend of distance bias f with distance s, proving a parallax offset delta p. On average, we find delta p = 0.054 mas (parallaxes in Gaia DR2 need to be increased) when accounting for the parallax uncertainty under-estimate in the Gaia set (delta p = 0.048 mas on the raw parallax errors) with negligible formal error and a systematic uncertainty of about 0.006 mas. The value is in concordance with results from asteroseismic measurements, but differs from the much lower bias found on quasar samples. We further use our method to compile a comprehensive set of quality cuts in colour, apparent magnitude, and astrometric parameters. Last, we find that for this sample delta p appears to strongly depend on the parallax uncertainty sigma_p (when including the additional 0.043 mas) with a statistical confidence far in excess of 10and a proportionality factor close to 1, though the dependence varies somewhat with sigma_p. Correcting for the sigma_p dependence also resolves otherwise unexplained correlations of the offset with the number of observation periods n_vis and ecliptic latitude.

The Software-defined Networking paradigm became widely accepted in academia as well as the industry in the last decade. The logically centralized control-plane provides simple mechanisms to support efficient monitoring as proposed in a plethora of works recently. However, existing works simplify the control-plane to a single entity whereas it is proven that the control-plane must be implemented physically distributed. To this end, we lately proposed D IST TM, a system to increase the efficiency while monitoring flows in distributed control-planes by eliminating redundant measurements of shared flows. The system is based on a centralized coordinator that introduces a single-point-of-failure and is vulnerable to network partitioning. To overcome this shortage, we propose a distributed algorithm to assign redundant measurement responsibilities to single con- trollers and share information among controllers in the work in hand. In an evaluation, we show that we decrease the number of measurements strongly and improve the load fairness among controllers through the collaboration of networks. Index Terms—Software-defined Networking, Flow Monitoring, Distributed Algorithms, Distributed Control Plane

In the practice of data analysis, there is a conceptual distinction between hypothesis testing, on the one hand, and estimation with quantified uncertainty on the other. Among frequentists in psychology, a shift of emphasis from hypothesis testing to estimation has been dubbed ``the New Statistics'' (Cumming 2014). A second conceptual distinction is between frequentist methods and Bayesian methods. Our main goal in this article is to explain how Bayesian methods achieve the goals of the New Statistics better than frequentist methods. The article reviews frequentist and Bayesian approaches to hypothesis testing and to estimation with confidence or credible intervals. The article also describes Bayesian approaches to meta-analysis, randomized controlled trials, and power analysis.

Few ideas have enjoyed as large an impact on deep learning as convolution. For any problem involving pixels or spatial representations, common intuition holds that convolutional neural networks may be appropriate. In this paper we show a striking counterexample to this intuition via the seemingly trivial coordinate transform problem, which simply requires learning a mapping between coordinates in (x,y) Cartesian space and one-hot pixel space. Although convolutional networks would seem appropriate for this task, we show that they fail spectacularly. We demonstrate and carefully analyze the failure first on a toy problem, at which point a simple fix becomes obvious. We call this solution CoordConv, which works by giving convolution access to its own input coordinates through the use of extra coordinate channels. Without sacrificing the computational and parametric efficiency of ordinary convolution, CoordConv allows networks to learn either complete translation invariance or varying degrees of translation dependence, as required by the end task. CoordConv solves the coordinate transform problem with perfect generalization and 150 times faster with 10--100 times fewer parameters than convolution. This stark contrast raises the question: to what extent has this inability of convolution persisted insidiously inside other tasks, subtly hampering performance from within? A complete answer to this question will require further investigation, but we show preliminary evidence that swapping convolution for CoordConv can improve models on a diverse set of tasks. Using CoordConv in a GAN produced less mode collapse as the transform between high-level spatial latents and pixels becomes easier to learn. A Faster R-CNN detection model trained on MNIST showed 24% better IOU when using CoordConv, and in the RL domain agents playing Atari games benefit significantly from the use of CoordConv layers.

Hybridization is an important force in evolution. The effects of hybridization across the whole genome are not understood. Using a fine-scale genetic map, Schumer et al. examined local ancestry in replicate natural hybrid populations of swordtail fish. Each parental species contributes different proportions of genetic material to the genomes of their descendants. Genes from the “minor” (less well-represented) parent occur in regions of the genome that are subject to higher recombination rates and where there are fewer potentially deleterious genes. Neanderthal ancestry in human genomes shows similar patterns.Science, this issue p. 656To investigate the consequences of hybridization between species, we studied three replicate hybrid populations that formed naturally between two swordtail fish species, estimating their fine-scale genetic map and inferring ancestry along the genomes of 690 individuals. In all three populations, ancestry from the “minor” parental species is more common in regions of high recombination and where there is linkage to fewer putative targets of selection. The same patterns are apparent in a reanalysis of human and archaic admixture. These results support models in which ancestry from the minor parental species is more likely to persist when rapidly uncoupled from alleles that are deleterious in hybrids. Our analyses further indicate that selection on swordtail hybrids stems predominantly from deleterious combinations of epistatically interacting alleles.

Networks and network architectures are constantly evolving, manifesting in new developments such as SDN, P4, and 400G Ethernet. These novel paradigms and technologies require network researchers to investigate and to adapt their measurement facilities. We present two tools which can foster this process. The first tool, pos, supports a fully automated workflow for performing and evaluating network experiments. One of its key features is the testbed orchestration to maintain and recreate a specified network test environment gaining reproducible experiment results. The second tool, libmoon, is a user-friendly userspace packet processing framework based on DPDK. Among many other projects, libmoon powers MoonGen, a dedicated packet generator in broad use by the community. MoonGen's hardware-supported generation and measurement capabilities are central to our network experiments to reliably recreate measurements. Further, a survey of scientific publications and applications presents projects based on both libmoon and MoonGen. We argue that combining pos and libmoon/MoonGen creates an ideal platform for network experiments. This platform offers an affordable price, high flexibility, ease of use and generation of reproducible experiments.

Abstract We examine the upper tail of flood peak distributions through analyses of annual peak observations from more than 8,000 U.S. Geological Survey (USGS) stream gaging stations and through hydrometeorological analyses of the storms that produce the most extreme floods. We focus on the distribution of the upper tail ratio, which is defined as the peak discharge for the flood of record at a stream gaging station divided by the sample 10-year flood magnitude. The 14 June 1903 Heppner storm, which produced an upper tail ratio of 200, was the product of a hailstorm that formed along the Blue Mountains in eastern Oregon, a region dominated by snowmelt flooding. A striking contrast between record flood peaks and the larger distribution of annual flood peaks in the United States is in the seasonality of flood occurrence, with record floods reflecting a much stronger contribution from warm season thunderstorm systems. Mountainous terrain and intense convective rainfall are important elements of the geography and hydrometeorology of extreme upper tail ratio flood peaks. The distribution of upper tail ratio values for USGS stream gaging stations does not depend on basin area, a result which is consistent with scaling results based on extreme value theory. Downscaling simulations with the Weather Research and Forecasting model are used to examine the storm environment of the 1903 Heppner storm, along with two other record flood peaks near the Blue Mountains of eastern Oregon from the USGS miscellaneous flood record, the July 1956 Meyers Canyon flood and the July 1965 Lane Canyon flood.

The concept of network slicing in 5G mobile networks introduces new challenges for security management: Given the combination of Infrastructure-as-a-Service cloud providers, mobile network operators as Software-as-a-Service providers, and the various verticals as customers, multi-layer and multi-tenancy-capable management architectures are required. This paper addresses the challenges for correlation of security events in such 5G scenarios with a focus on event processing at telecommunication service providers. After an analysis of the specific demand for network-slice-centric security event correlation in 5G networks, ongoing standardization efforts, and related research, we propose a multi-tenancy-capable event correlation architecture along with a scalable information model. The event processing, alerting, and correlation workflow is discussed and has been implemented in a network and security management system prototype, leading to a demonstration of first results acquired in a lab setup.

The correct identification of differentially expressed genes (DEGs) between specific conditions is a key in the understanding phenotypic variation. High-throughput transcriptome sequencing (RNA-Seq) has become the main option for these studies. Thus, the number of methods and softwares for differential expression analysis from RNA-Seq data also increased rapidly. However, there is no consensus about the most appropriate pipeline or protocol for identifying differentially expressed genes from RNA-Seq data. This work presents an extended review on the topic that includes the evaluation of six methods of mapping reads, including pseudo-alignment and quasi-mapping and nine methods of differential expression analysis from RNA-Seq data. The adopted methods were evaluated based on real RNA-Seq data, using qRT-PCR data as reference (gold-standard). As part of the results, we developed a software that performs all the analysis presented in this work, which is freely available at https://github.com/costasilvati/consexpression. The results indicated that mapping methods have minimal impact on the final DEGs analysis, considering that adopted data have an annotated reference genome. Regarding the adopted experimental model, the DEGs identification methods that have more consistent results were the limma+voom, NOIseq and DESeq2. Additionally, the consensus among five DEGs identification methods guarantees a list of DEGs with great accuracy, indicating that the combination of different methods can produce more suitable results. The consensus option is also included for use in the available software.

t-distributed Stochastic Neighborhood Embedding (t-SNE), a clustering and visualization method proposed by van der Maaten & Hinton in 2008, has rapidly become a standard tool in a number of natural sciences. Despite its overwhelming success, there is a distinct lack of mathematical foundations and the inner workings of the algorithm are not well understood. The purpose of this paper is to prove that t-SNE is able to recover well-separated clusters; more precisely, we prove that t-SNE in the `early exaggeration' phase, an optimization technique proposed by van der Maaten & Hinton (2008) and van der Maaten (2014), can be rigorously analyzed. As a byproduct, the proof suggests novel ways for setting the exaggeration parameter $\alpha$ and step size $h$. Numerical examples illustrate the effectiveness of these rules: in particular, the quality of embedding of topological structures (e.g. the swiss roll) improves. We also discuss a connection to spectral clustering methods.

Für die Funktion des IT-Produktmanagements können bislang weder in Wissenschaft noch Praxis ausreichende Erkenntnisse aufgewiesen und zudem kein anerkanntes Modell verzeichnet werden, um IT-Produktmanager unterschiedlicher Organisationen in die Lage zu versetzen, die jeweiligen IT-Produkte zum Erfolg zu führen. Diese Arbeit verfolgt das gestaltungsorientierte Ziel, ein theoriegeleitetes, hypothetisches, validiertes, situatives Modell für das IT-Produktmanagement als Ergebnis aus der Zusammenführung von Theorie und Praxis zu entwickeln, das zusätzlich Gestaltungsempfehlungen für IT-Produktmanager geben kann. Hierbei wird besonderes Augenmerk auf gemeinsame Merkmale gelegt, die eine Typisierung von IT-Produktmanagern zulassen. Um dieses Ziel zu erreichen, wird die Stelle des IT-Produktmanagements im situativen Kontext aus unterschiedlichen Perspektiven erforscht und konkretisiert. Aufbauend auf den wissenschaftlichen Erkenntnissen und den empirischen Befunden aus der Praxis, wird durch sensibilisierende Konzepte ein theoriegeleitetes Modell für das IT-Produktmanagement (SIMOIT-PM) erstellt. Identifiziert werden sechs Referenztypen. Durch eine Zuordnung zu diesen Typen können IT-Produktmanager durch Ableitung von Gestaltungsempfehlungen bei der Betreuung ihrer IT-Produkte unterstützt werden. Zuletzt erfolgt eine Validierung, zum einen argumentativ, um die Rigorosität der Ergebnisse darzustellen, zum anderen an der Praxis durch Fallstudien, um die Anwendbarkeit und Funktionalität des SIMOIT-PM aufzuzeigen.

Two-ray ground reflection model has been widely used as the propagation model to investigate the performance of an ad hoc network. But two-ray model is too simple to represent a real world network. A more realistic model namely shadowing propagation model has been used in this investigation. Under shadowing propagation model, a mobile node may receive a packet at a signal level that is below a required threshold level. This low signal level affects the routing protocol as well as the medium access control protocol of a network. An analytical model has been presented in this paper to investigate the shadowing effects on the network performance. The analytical model has been verified via simulation results. Simulation results show that the performance of a network becomes very poor if shadowing propagation model is used in compare to the simple two-ray model. Two solutions have also been proposed in this paper to overcome the effects of shadowing. One solution is a physical layer solution and the other one is a Medium Access Control (MAC) layer solution. Simulation results show that these two solutions reduce the shadowing effect and improve network performance.