TY - GEN ED - COMPASS '18, T1 - Two Approximate Dynamic Programming Algorithms for Managing Complete SIS Networks N2 - Inspired by the problem of best managing the invasive mosquito Aedes albopictus across the 17 Torres Straits islands of Australia, we aim at solving a Markov decision process on large Susceptible-Infected-Susceptible (SIS) networks that are highly connected. While dynamic programming approaches can solve sequential decision-making problems on sparsely connected networks, these approaches are intractable for highly connected networks. Inspired by our case study, we focus on problems where the probability of nodes changing state is low and propose two approximate dynamic programming approaches. The first approach is a modified version of value iteration where only those future states that are similar to the current state are accounted for. The second approach models the state space as continuous instead of binary, with an on-line algorithm that takes advantage of Bellman's adapted equation. We evaluate the resulting policies through simulations and provide a priority order to manage the 17 infested Torres Strait islands. Both algorithms show promise, with the continuous state approach being able to scale up to high dimensionality (50 nodes). This work provides a successful example of how AI algorithms can be designed to tackle challenging computational sustainability problems. Y1 - 2018 SN - 978-1-4503-5816-3 U6 - https://doi.org/10.1145/3209811.3209814 PB - ACM New York, NY, USA ER - TY - JOUR A1 - Hassan, Irtaza A1 - Donati, Luca A1 - Stensitzki, Till A1 - Keller, Bettina G A1 - Heyne, Karsten A1 - Imhof, Petra T1 - The vibrational spectrum of the hydrated alanine-leucine peptide in the amide region from IR experiments and first principles calculations JF - Chemical Physics Letters N2 - We have combined infrared (IR) experiments with molecular dynamics (MD) simulations in solution at finite temperature to analyse the vibrational signature of the small floppy peptide Alanine-Leucine. IR spectra computed from first-principles MD simulations exhibit no distinct differences between conformational clusters of -helix or -sheet-like folds with different orientations of the bulky leucine side chain. All computed spectra show two prominent bands, in good agreement with the experiment, that are assigned to the stretch vibrations of the carbonyl and carboxyl group, respectively. Variations in band widths and exact maxima are likely due to small fluctuations in the backbone torsion angles. Y1 - 2018 U6 - https://doi.org/10.1016/j.cplett.2018.03.026 VL - 698 SP - 227 EP - 233 ER - TY - CHAP A1 - Keller, Bettina G. A1 - Aleksić, Stevan A1 - Donati, Luca T1 - Markov State Models in Drug Design T2 - Biomolecular Simulations in Structure‐Based Drug Discovery N2 - This chapter explains the different ways in which Markov State Models (MSMs) can be helpful in structure-based drug design. MSMs are constructed from the time series of molecular dynamics (MD), which can be generated by classical MD simulations. Several features of the MSMs can be utilized for rational drug design. The discretization of a validated MSM is particularly suited to extract meaningful representatives from the conformational ensemble, because the discretization yields a small number of microstates and mirrors the features of the free energy landscape. Long-lived conformations consist of a set of microstates which show high transition rates within the set and low transition rates to microstates outside of the set. The Bayesian agglomerative clustering engine (BACE) algorithm uses the observed transition counts to extract long-lived conformations from an MSM. By iteratively merging microstates according to the Bayes factor and recalculating the Bayes-factor matrix, the algorithm yields an aggregation of the microstates into long-lived conformations. Y1 - 2018 SN - 1865-0562 U6 - https://doi.org/10.1002/9783527806836.ch4 SP - 67 EP - 86 PB - Wiley-Interscience, Weinheim ER - TY - JOUR A1 - Donati, Luca A1 - Keller, Bettina G. T1 - Girsanov reweighting for metadynamics simulations JF - The Journal of Chemical Physics N2 - Metadynamics is a computational method to explore the phase space of a molecular system. Gaussian functions are added along relevant coordinates on the fly during a molecular-dynamics simulation to force the system to escape from minima in the potential energy function. The dynamics in the resulting trajectory are however unphysical and cannot be used directly to estimate dynamical properties of the system. Girsanov reweighting is a recent method used to construct the Markov State Model (MSM) of a system subjected to an external perturbation. With the combination of these two techniques—metadynamics/Girsanov-reweighting—the unphysical dynamics in a metadynamics simulation can be reweighted to obtain the MSM of the unbiased system. We demonstrate the method on a one-dimensional diffusion process, alanine dipeptide, and the hexapeptide Val-Gly-Val-Ala-Pro-Gly (VGVAPG). The results are in excellent agreement with the MSMs obtained from direct unbiased simulations of these systems. We also apply metadynamics/Girsanov-reweighting to a β-hairpin peptide, whose dynamics is too slow to efficiently explore its phase space by direct simulation Y1 - 2018 U6 - https://doi.org/10.1063/1.5027728 VL - 149 IS - 7 SP - 072335 ER - TY - JOUR A1 - Wakayama, Yuki A1 - Yokoyama, Ryohei A1 - Shinano, Yuji A1 - Wakui, Tetsuya T1 - エネルギー供給システムの設計および運用の階層的関係を考慮した最適化(時間集約によるモデル縮約とその効果) T1 - Optimization of Energy Supply Systems in Consideration of Hierarchical Relationship Between Design and Operation(Model Reduction by Time Aggregation and Its Effect) JF - エネルギー・資源学会第34回エネルギーシステム・経済・環境コンファレンス講演論文集 N2 - A hierarchical mixed-integer linear programming method to solve optimal design problems of energy supply systems efficiently has been proposed. The original problem is solved by dividing it into a relaxed optimal design problem at the upper level and optimal operation problems which are independent of each other at the lower level. In this paper, a method of reducing model by time aggregation is proposed to search design solution candidates efficiently in the relaxed optimal design problem at the upper level. This method is realized only by clustering periods and averaging energy demands for clustered periods, while it guarantees to derive the optimal solution. On one hand, the method may decrease the number of design variables and costraints at the upper level, and thus the computation time at the upper level. On the other hand, it may increase the numbers of generated design solution candidates and solved optimal operation problems, and thus the computation time at both the levels. Though a case study on the optimal design of a cogeneration system, it is clarified how the model reduction is effective to enhance the computation efficiency. Y1 - 2018 SP - 401 EP - 404 ER - TY - JOUR A1 - Wakayama, Yuki A1 - Yokoyama, Ryohei A1 - Shinano, Yuji A1 - Wakui, Tetsuya T1 - 階層的最適化によるエネルギ ー供給システムの最適設計・運用 (期間のクラスタリングに よるモデル縮約とその効果) T1 - Optimal Design and Operation of Energy Supply Systems by Hierarchical Optimization (Model Reduction by Clustering Periods and Its Effect) JF - 日本機械学会関西支部第93期定時総会講演会講演論文集 N2 - Ahierarchical mixed-integer linear programmingmethod has been proposed to solve optimal design problems of energy supply systems efficiently. In this paper, a method of reducing model by clustering periods is proposed to search design solution candidates efficiently in the relaxed optimal design problem at the upper level. This method is realized only by clustering periods and averaging energy demands for clustered periods, while it guarantees to derive the optimal solution. Through acase study on the optimaldesign of a cogeneration system, ti is clarified how the model reduction si effective ot enhance the computation efficiency. Y1 - 2018 SP - 259 EP - 260 ER - TY - CHAP A1 - Kramer, Tobias A1 - Läuter, Matthias T1 - Modelling the inner coma of comet 67P/Churyumov-Gerasimenko T2 - Europlanet Science Congress N2 - Based on about 1 million of pressure measurements around comet 67P/Churyumov-Gerasimenko we reconstruct the gas emission across the entire nucleus. Dust particles are seeded in the gas model and the resulting dust distribution follows a daily pattern which agrees with observations if a uniform dust release across the entire sunlit surface is assumed. Y1 - 2018 VL - 12 SP - EPSC2018-115-1 PB - Europlanet Science Congress ER - TY - CHAP A1 - Läuter, Matthias A1 - Kramer, Tobias A1 - Rubin, Martin A1 - Altwegg, Kathrin T1 - Gas production of comet 67P/Churyumov-Gerasimenko reconstructed from DFMS/COPS data T2 - Europlanet Science Congress N2 - We reconstruct the temporal evolution of surface emissions for the four major gas species H2O, CO2, CO, and O2 emitted during the 2015 apparition of comet 67P/Churyumov-Gerasimenko (67P/C-G). Measurements from the Double Focusing Mass Spectrometer (DFMS) of the Rosetta Orbiter Spectrometer for Ion and Neutral Analysis (ROSINA) and the COmet Pressure Sensor (COPS) are used to determine the gas sources on the surface with an inverse gas model for the entire coma. For all species, peak production rates and integrated production rates per orbit are evaluated separately for the northern and the southern hemisphere. Complemented with the total mass production, this allows us to estimate the dust-to-gas ratio of the emitted material. Y1 - 2018 VL - 12 SP - EPSC2018-515-1 PB - Europlanet Science Congress ER - TY - JOUR A1 - Bonami, Pierre A1 - Gleixner, Ambros A1 - Linderoth, Jeff A1 - Misener, Ruth T1 - Designing and Implementing Algorithms for Mixed-Integer Nonlinear Optimization (Dagstuhl Seminar 18081) JF - Dagstuhl Reports N2 - Mathematical models for optimal decisions often require both nonlinear and discrete components. These mixed-integer nonlinear programs (MINLP) may be used to optimize the energy use of large industrial plants, integrate renewable sources into energy networks, design biological and biomedical systems, and address numerous other applications of societal importance. The first MINLP algorithms and software were designed by application engineers. While these efforts initially proved useful, scientists, engineers, and practitioners have realized that a transformational shift in technology will be required for MINLP to achieve its full potential. MINLP has transitioned to a forefront position in computer science, with researchers actively developing MINLP theory, algorithms, and implementations. Even with their concerted effort, algorithms and available software are often unable to solve practically-sized instances of these important models. Current obstacles include characterizing the computability boundary, effectively exploiting known optimization technologies for specialized classes of MINLP, and effectively using logical formulas holistically throughout algorithms. Y1 - 2018 U6 - https://doi.org/10.4230/DagRep.8.2.64 VL - 8 IS - 2 SP - 64 EP - 87 ER - TY - JOUR A1 - Bärmann, Andreas A1 - Martin, Alexander A1 - Pokutta, Sebastian A1 - Schneider, Oskar T1 - An Online-Learning Approach to Inverse Optimization Y1 - 2018 ER - TY - JOUR A1 - Bienenstock, Daniel A1 - Muñoz, Gonzalo A1 - Pokutta, Sebastian T1 - Principled Deep Neural Network Training through Linear Programming N2 - Deep Learning has received significant attention due to its impressive performance in many state-of-the-art learning tasks. Unfortunately, while very powerful, Deep Learning is not well understood theoretically and in particular only recently results for the complexity of training deep neural networks have been obtained. In this work we show that large classes of deep neural networks with various architectures (e.g., DNNs, CNNs, Binary Neural Networks, and ResNets), activation functions (e.g., ReLUs and leaky ReLUs), and loss functions (e.g., Hinge loss, Euclidean loss, etc) can be trained to near optimality with desired target accuracy using linear programming in time that is exponential in the input data and parameter space dimension and polynomial in the size of the data set; improvements of the dependence in the input dimension are known to be unlikely assuming P≠NP, and improving the dependence on the parameter space dimension remains open. In particular, we obtain polynomial time algorithms for training for a given fixed network architecture. Our work applies more broadly to empirical risk minimization problems which allows us to generalize various previous results and obtain new complexity results for previously unstudied architectures in the proper learning setting. Y1 - 2018 ER - TY - THES A1 - Dill, Sabrina T1 - Joint Feature Learning and Classification - Deep Learning for Surgical Phase Detection N2 - In this thesis we investigate the task of automatically detecting phases in surgical workflow in endoscopic video data. For this, we employ deep learning approaches that solely rely on frame-wise visual information, instead of using additional signals or handcrafted features. While previous work has mainly focused on tool presence and temporal information for this task, we reason that additional global information about the context of a frame might benefit the phase detection task. We propose novel deep learning architectures: a convolutional neural network (CNN) based model for the tool detection task only, called Clf-Net, as well as a model which performs joint (context) feature learning and tool classification to incorporate information about the context, which we name Context-Clf-Net. For the phase detection task lower-dimensional feature vectors are extracted, which are used as input to recurrent neural networks in order to enforce temporal constraints. We compare the performance of an online model, which only considers previous frames up to the current time step, to that of an offline model that has access to past and future information. Experimental results indicate that the tool detection task benefits strongly from the introduction of context information, as we outperform both Clf-Net results and stateof-the-art methods. Regarding the phase detection task our results do not surpass state-of-the-art methods. Furthermore, no improvement of using features learned by the Context-Clf-Net is observed in the phase detection task for both online and offline versions Y1 - 2018 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:0297-zib-81745 ER - TY - THES A1 - Schotte, Merlind T1 - Finite-Elemente-Techniken zur mechanischen Stabilitätsanalyse der Hyomandibula von Rochen N2 - Haie und Rochen sind die einzigen Wirbeltiere, die sich durch ein Knorpelskelett auszeichnen, das mit winzigen mineralisierten Elementen bedeckt ist, die sogenannten Tesserae. Obwohl dieser mineralisierte Knorpel seit hunderten von Millionen Jahren ein charakteristisches Merkmal dieser Tiere ist, ist der funktionelle Vorteil dieses Gewebestruktur bisher unklar geblieben. In dieser Masterarbeit wird diese Struktur mithilfe der Finite-Elemente-Methode(FEM) untersucht, wobei biologische Informationen zu den Muskeln und der geometrischen Gegebenheiten aus einem hochaufgelösten, kontrastgefärbten μCT-Scan einer Stechrochenhyomandibula verwendet werden. Für die Analyse werden Computer-Aided-Design(CAD)-Modelle und FE-Modelle konstruiert sowie vorhandene Methoden erweitert, um bei den Berechnungen die Charakteristika des biologischen Objektes bestmöglich zu berücksichtigen. Die Modelle bauen auf einer Segmentierung der Hyomandibula auf, wobei die mehreren tausend Tesserae in diesem Datensatz bereits voneinander isoliert sind. Für die Modellierung der Hyomandibula wird als erstes ein CAD-Modell konstruiert. Bei dieser Konstruktion wird eine Tesserae-Mittelfläche erstellt und im Anschluss nach innen und außen verschoben. Der nächste Schritt basiert auf dem Einbau von Zwischenwänden, die die Tesserae voneinander separieren. Auf Grundlage dieses Modells wird ein Tetraedergitter generiert, das die Basis für den zweiten Teil dieser Arbeit darstellt - die Berechnungen mit der FEM. Die wesentliche Aufgabe des zweiten Teils beinhaltet die Konstruktion eines FE-Modells, das in die lineare FEM einzuordnen ist. In dieses fließen die Randbedingungen und die Materialeigenschaften des Skelettelements ein. Mithilfe der FEM lassen sich die durch Krafteinwirkung auftretenden Verschiebungen und Spannungen der Hyomandibula ermitteln. Die Ergebnisse geben Rückschlüsse über das Verhalten des Materials, die Besonderheiten der Geometrie und damit über die Stabilität des Objektes. Der dritte Teil der Abschlussarbeit beschäftigt sich mit zusätzlichen Anpassungen und Änderungen der Modelle (CAD-Modell und FE-Modell), um diese Resultate mit dem Modell zu vergleichen, das der Biologie am nächsten kommt. Diese Anpassungen beinhalten das Variieren der Dicke der mineralisierten Schicht, die Konstellation des Materials und den Abstand zwischen den Tesserae. Die Ergebnisse zeigen, dass sowohl die Tesserae-Struktur als auch die zwischen den Tesserae liegenden Kollagenfasern vermutlich keinen Einfluss auf die Mechanik und dementsprechend die Stabilität haben. Jedoch lassen die unterschiedlichen Resultate im Fall verschieden gewählter Tesserae-Dicken annehmen, dass diese sehr wohl einen Einfluss besitzen. N2 - Sharks and rays are the only vertebrates characterized by a peculiar cartilage skeleton covered with minute mineralized tiles called tesserae. Although this tessellated cartilage has been a defining feature of this lineage for hundreds of millions of years, the functional advantage of this tissue design has remained obscure. In this Master’s thesis, the finite element method (FEM) is used to investigate the role of tiling in the skeleton, using biological information on the muscles and shape from a high-resolution, contrast-stained μCT-scan of a stingray hyomandibula. For this purpose, computer aided design (CAD) models and FE models are constructed and existing methods will be extended to take the best possible account of the characteristics of the biological object. The analysis builds from a segmentation of the hyomandibula, with the dataset’s several thousand tesserae pre-isolated from one another. For the modeling of the hyomandibula a CAD model is constructed. In this construction, a medial surface is created using the tesserae which is subsequently moved to the inside and outside. The next step is to integrate partition walls separating the tesserae elements. Based on this model, a tetrahedral grid is then generated, which is the content of the second part of this thesis, the calculation with the FEM. A finite element model is constructed including boundary conditions and material properties of the skeletal element to compute the results using the linear FEM theory. The FEM is used to determine the displacements and stresses resulting from biological forces acting on the hyomandibula. The results allow conclusions of the material behavior, geometric characteristics (tesselation) and therefore the object stability. Important contributions of this thesis are additional adjustments and changes to the models (CAD model and FE model) to make it more appropriate to study the biological data at hand. The tesserae thickness, the length of the collagenious fibres and the constellation of the material are the parameters, which are changed. In contrast to the tesserae thicknesses, the results suggest that both the arrangement of the tesserae and its interposed collagen fibers have no substantial influence on the mechanics. Y1 - 2018 ER - TY - THES A1 - Weltsch, André T1 - Fast Approximation of Equations of transient Gasflow Y1 - 2018 ER - TY - CHAP A1 - Inanlouganji, A. A1 - Pedrielli, G. A1 - Fainekos, G. A1 - Pokutta, Sebastian T1 - Continuous Simulation Optimization with Model Mismatch Using Gaussian Process Regression T2 - Proceedings of the 2018 Winter Simulation Conference Y1 - 2018 ER - TY - CHAP A1 - Pokutta, Sebastian A1 - Singh, M. A1 - Torrico, A. T1 - Efficient algorithms for robust submodular maximization under matroid constraints T2 - ICML Workshop Paper Y1 - 2018 N1 - Additional Note: https://arxiv.org/abs/1807.09405 N1 - URL of the PDF: https://sites.google.com/view/icml2018nonconvex/papers ER - TY - JOUR A1 - Le Bodic, P. A1 - Pfetsch, Marc A1 - Pavelka, J. A1 - Pokutta, Sebastian T1 - Solving MIPs via Scaling-based Augmentation JF - Discrete Optimization Y1 - 2018 N1 - Additional Note: doi: 10.1016/j.disopt.2017.08.004 N1 - URL of the PDF: http://dx.doi.org/10.1016/j.disopt.2017.08.004 VL - 27 SP - 1 EP - 25 ER - TY - JOUR A1 - Braun, Gábor A1 - Roy, Aurko A1 - Pokutta, Sebastian T1 - Stronger Reductions for Extended Formulations JF - to appear in Mathematical Programming B Y1 - 2018 N1 - Additional Note: available at \url{http://arxiv.org/abs/1512.04932} ER - TY - JOUR A1 - Song, R. A1 - Xie, Y. A1 - Pokutta, Sebastian T1 - On the effect of model mismatch for sequential Info-Greedy Sensing JF - EURASIP Journal on Advances in Signal Processing Y1 - 2018 N1 - URL of the PDF: https://asp-eurasipjournals.springeropen.com/articles/10.1186/s13634-018-0551-y ER - TY - CHAP A1 - Paetsch, Olaf T1 - Möglichkeiten und Grenzen automatischer Merkmalserkennung am Beispiel von Risserkennungen in 3D-CT-Aufnahmen von Betonproben N2 - Die Auswertungen der großen Datenmengen moderner bildgebender Verfahren der ZfP können manuell kaum noch bewältigt werden. Hochauflösende 3D-CT-Aufnahmen bestehen oft aus über 1000 Schichtbildern mit einer Datenmenge von mehreren Gigabytes. Aktuelle Computer können diese zwar problemlos visualisieren und erlauben somit eine visuelle Inspektion, aber die möglichst vollständige Erkennung bestimmter Merkmale in den Daten und deren qualitative wie quantitative Auswertung ist durch Experten manuell nicht mehr zu bewältigen. Das gilt insbesondere im Kontext der Schadensaufklärung für die quantitative Analyse verschiedenartig induzierter Risse in Betonen (z.B. durch mechanische Belastungen sowie Frost, Sulfat und Alkali-Kieselsäure-Reaktion). Eine dazu notwendige Segmentierung und Merkmalserkennung kann nur automatisch durchgeführt werden. Dabei ergibt sich (auch fast automatisch) die Frage nach der Verlässlichkeit der verwendeten Algorithmen. Inwieweit kann man davon ausgehen, dass alle gesuchten Merkmale auch tatsächlich gefunden worden sind? Sind die gefundenen Merkmale quantitativ auswertbar und wie wirken sich Parameteränderungen auf die Ergebnisse aus? Sollten immer dieselben, einmal mit gutem Ergebnis angewandten, Parameter auch bei anderen Proben zur besseren Vergleichbarkeit genutzt werden? Anhand eines Risserkennungsalgorithmus basierend auf Formerkennung und Bildverarbeitung wird die Problematik diskutiert. Als Grundlage zur Bewertung des Ansatzes dienen 3D-CT-Aufnahmen von geschädigten Betonprobekörpern und Datensätze, in denen sich aufgrund ihres homogenen Aufbaus mit einfachen Mitteln Risse sicher und eindeutig erkennen lassen. Zur Auswertung der erkannten Risse gehört auch deren Einbettung in das umliegende Material. Das erfordert neben einer automatischen Risserkennung auch eine Segmentierung des gesamten Probekörpers in Zementsteinmatrix, Gesteinskörnung und Porenraum. Da eine solche Segmentierung aufgrund der Datenmenge nur schwer manuell erfolgen kann, werden erste Ergebnisse aus einer Segmentierung mit Hilfe selbstlernender Convolutional Neural Networks gezeigt. Y1 - 2018 PB - iCT 2019 CY - https://www.ndt.net/article/ctc2019/papers/iCT2019_Full_paper_29.pdf ER -