ENBIS-17 in Naples

9 – 14 September 2017; Naples (Italy) Abstract submission: 21 November 2016 – 10 May 2017

My abstracts


The following abstracts have been accepted for this event:

  • Statistical Methods and Tools as Drivers of Innovation in the Energy Industry: Some Insights and Examples

    Authors: Alberto Pasanisi (EDF R&D - EIFER)
    Primary area of focus / application: Other: ENBIS Best manager Award
    Keywords: Statistics, R&D, Innovation, Energy, Modelling, Simulation
    Submitted at 9-Feb-2017 14:01 by Alberto Pasanisi
    Modern engineering and business require undoubtedly more and more multidisciplinary skills. As a heavy trend in the current practice of engineering, it is worth highlighting the increasing use of computer simulation and the exploitation of more and more huge and heterogeneous sources of data. In the end, today’s engineers need to complete and to update constantly their toolbox with tools coming from the domains of information technology and applied mathematics.
    In particular, statistics and data analysis provide valuable methods and tools to quantify and manage uncertainties when forecasting the behaviour of industrial or natural systems, to extract knowledge from data and expertise and to recommend decisions: actually, they help solving a consistent part of problems, today’s engineering is concerned with.
    Inspired from the feedback of more than 15 years in the field of advanced engineering and R&D, this talk highlights some examples of the use of statistical methods to implement innovative solutions for industrial problems, with a particular focus on the energy industry.
    Without any pretention of exhaustiveness, the examples cover a great variety of energy-related activities (reliability of components and structures, natural hazards, energy efficiency, smart and sustainable cities ...), and put into evidence the cross-disciplinary and crucial role played by statistical methods.
  • Strategies for Mixture-Design Space Augmentation

    Authors: Pat Whitcomb (Stat-Ease, Inc.), Martin Bezener (Stat-Ease, Inc.)
    Primary area of focus / application: Design and analysis of experiments
    Secondary area of focus / application: Design and analysis of experiments
    Keywords: Design of Experiments, Mixture design, Augmentation, RSM
    Submitted at 17-Feb-2017 21:42 by Pat Whitcomb
    Accepted (view paper)
    12-Sep-2017 10:50 Strategies for Mixture-Design Space Augmentation
    One of the critical steps during the planning of a mixture experiment is the selection of the component lower and upper bounds. These bounds should be chosen wide enough such that an interesting change in the response occurs somewhere in between the lows and highs, but not too wide as to create extreme conditions. Ideally, any optimum should occur near the center of the experimental region, where prediction precision is high. In practice, however, at least one of these bounds is often chosen to be too narrow, causing an optimum to occur along an edge or vertex of the experiment-design space. In our experience, industrial experimenters will often simply re-run an entirely new experiment with wider ranges due to lack of readily-available augmentation strategies.

    Augmentation and sequential strategies for response surface methodology (RSM) have been typically studied in the context of model order or sample size. However, strategies for the expansion of the region for experimentation are sparse, especially for mixture designs. In this talk, we briefly describe the problem at hand for RSM in general. Then we pay particular attention to mixture experiments, where expanding component ranges is more complicated due to their interdependence on one another. We propose several strategies, including an optimal DOE space augmentation algorithm. Several examples will be given.
  • Updating Monitoring Networks in Overlay Data Modelling

    Authors: Riccardo Borgoni (University of Milano Bicocca)
    Primary area of focus / application: Other: Monitoring and optimization of semiconductor processes
    Keywords: Microelectronics, Overlay, Lithography, Optimal design, Spatial network reduction, Metaheuristic optimization
    Submitted at 20-Feb-2017 15:36 by Riccardo Borgoni
    11-Sep-2017 18:10 Updating Monitoring Networks in Overlay Data Modelling
    Integrated circuits are built by a sequence of patterning steps that form subsequent layers onto a semiconductor wafer. The pattern created at each step should be aligned to pre-existing patterns. The difference between the current layer position and the substrate geometry is called overlay. The wafer area is divided in portions called fields that are scanned sequentially to measure overlay displacements. In order to estimate models for adjusting overlay in subsequent steps, data are collected using a network of monitoring points, called target points, located at the border of the fields. However, measuring procedures are time consuming and expensive, hence, it is worth trying to reduce the number of points that are necessary to evaluate accurately the displacement in order to speed up the fabrication process.
    In this paper, we propose a maxent and a maxmin strategy, based on the tree spanning the sampling points, to select an optimal subsample of a given size of the network. The objective function measures how much the network is spread to cover the wafer area. Inspecting all the possible configurations of n points out of the N points originally present in the network is practically unfeasible when the number of measurement locations is even moderately high, hence metaheuristic optimization is employed to tackle this problem.
    We compared the results obtained using the reduced network to those obtained using the full sample in terms of the precision of both the predicted overlay values and the estimates of the regression coefficients of the calibrating model. It has been found, that even halving the sample size, the performance of the reduced network remains substantially unchanged.
  • From Life Tables to Deep Learning: Where Is the Insurance Business Going? What Are the Opportunities and Challenges for Statisticians?

    Authors: Yves Grize (Baloise Insurance & ZHAW)
    Primary area of focus / application: Other: ISBIS Session on Statistics in Economics and Business
    Keywords: Statistics in insurance, Digitalization, Deep Learning, Big Data, Actuarial sciences, Predictive Analytics, Statistics in industry
    Submitted at 20-Feb-2017 16:55 by Yves Grize
    Accepted (view paper)
    11-Sep-2017 12:00 From Life Tables to Deep Learning: Where Is the Insurance Business Going? What Are the Opportunities and Challenges for Statisticians?
    Once again is the insurance industry confronted with a major revolution: after the deregulation of the industry about 20 years ago, it is facing today a "digital revolution".
    Deregulation brought many exciting opportunities to the field of statistics, as it became clear that statistical know-how (later called predictive analytics), was key to profitability in the new competitive environment. The second revolution is coming now from the technological side: digitalization, big data, cognitive computing are some of the buzzwords here. However this time, the disruptions are likely to affect in-depth how insurers deal with data and risks. Some even predict that this revolution may change the entire business model of insurance companies.

    I will briefly review the potential impact of this second revolution on the insurance business in general and try to see in particular, what challenges and opportunities may result for the actuaries & statisticians of the XXI century.
  • Four Algorithms to Construct a Sparse Kriging Kernel for Dimensionality Reduction

    Authors: Mélina Ribaud (Ecole Centrale de Lyon)
    Primary area of focus / application: Modelling
    Secondary area of focus / application: Design and analysis of experiments
    Keywords: Kriging, High dimension, Covariance kernel, Isotropic, Anisotropic, Algorithms
    Submitted at 21-Feb-2017 10:26 by Mélina Ribaud
    12-Sep-2017 10:10 Four Algorithms to Construct a Sparse Kriging Kernel for Dimensionality Reduction
    We propose a new kriging kernel adapted to high dimension problems. Usually, people use an anisotropic kernel that depends on as many range parameters as the number of inputs. In high dimension with a very restricted number of data points the parameters estimation becomes difficult.
    An isotropic kernel parameterized by only one range parameter could be an alternative but it is too restrictive. The idea is to construct a kernel between these two extremal choices.
    We propose a data-driven construction of a covariance kernel adapted to high dimension. This kernel is a tensor product of isotropic kernels built on well-chosen subgroups of variables. Four algorithms are implemented to find the number of groups and their composition. They start with an isotropic and finish with an isotropic by group. At each stage, different models are compared and the best is chosen under a BIC criterion.
    The first algorithm explores only in the forward direction whereas the three others move in both directions. In the third and last algorithm a cluster step is added at different stage to improve the flexibility.
    The algorithms are compared on a simulated case in dimension eight. The fourth algorithm is the most efficient regarding to the set size, the prediction quality and the running time.The isotropic by group kernel is benchmarked against classical kernels on the Sobol function. The study shows that the isotropic by group kernel improves the prediction power.
  • Simulating High Correlations Using Ensemble Copula Coupling Method

    Authors: Jérôme Collet (EDF R&D)
    Primary area of focus / application: Other: Modeling, forecasting and risk evaluation of wind energy production
    Keywords: Ensemble copula coupling, Density forecast, Risk management, Correlation
    Submitted at 21-Feb-2017 14:11 by Jérôme Collet
    Accepted (view paper)
    12-Sep-2017 10:30 Simulating High Correlations Using Ensemble Copula Coupling Method
    Ensemble Copula Coupling method is now widely used for probability forecasting, which is compulsory for renewables management. We show here that, if using random samples, it is not possible to reproduce a very high correlation with Ensemble Copula Coupling method. If the observations are highly correlated, the simulations will be a bit less correlated. This issue has been yet signalled by Golestaneh, Beng Gooi, and Pinson, but its cause was not clearly stated. We show here that the only cause is the value of the correlation, we compute analytically the bias in some cases, show low sensitivity to distribution form, and propose generic solutions.