ENBIS-13 in Ankara

15 – 19 September 2013 Abstract submission: 5 February – 5 June 2013

My abstracts

 

The following abstracts have been accepted for this event:

  • Multi-objective Optimization of Contingency Logistics Networks with Distorted Risks

    Authors: Esra Dağ (Toros Üniversitesi), Mehmet Miman (Toros Üniversitesi)
    Primary area of focus / application: Reliability
    Keywords: Contingency logistics networks, Multi-criteria optimization, Distorted risks, Reliability
    Submitted at 3-Apr-2013 14:29 by Esra Dağ
    Accepted
    18-Sep-2013 09:40 Multi-objective Optimization of Contingency Logistics Networks with Distorted Risks
    Contingencies are unexpected crises that cause a major threat to the safety and well being of a specific population. To perform specific ontingency operations bases are holding stocks. The reliability of base is defined based on interference theory between supply and demand carried to and emerged from the base. In a contingency setting demand emerged from operational base and supply that can be carried to the base are random and unknown beforehand. If the supply to the base is less than demand emerged the base is called as “failed” as the base in incapable of performing the associated contingency operation. The decision maker’s tolerance for the risk associated with a node (base) failure is incorporated through the use of distortion. Distortion amplfies the perceived risks associated with node failure depending on the decision maker's attitudes (risk averse or risk seeker) towrds the risk. Specifically, the dual power and proportional hazard distortion models are investigated in this modeling paradigm.
    In our study we are exploring affect of distortion in multi-objective optimization setting where criterias are cost, network reliability and allocated stocks. The weighted objectives methods is used as a multi-criteria optimization technique. Our study is exploring the effects of distortion in multi-criteria optimization of the contingency logistic networks (CLN) for the first time.
  • Initial Results from Herd Behavior Models in Social Networks

    Authors: Alon Sela (Tel Aviv University), Irad Ben-Gal (Tel Aviv University)
    Primary area of focus / application: Mining
    Keywords: Herd behaviour, Social networks, Information flow
    Submitted at 3-Apr-2013 15:29 by Irad Ben-Gal
    Accepted
    This work considers the flow of information through social networks as consequence of herd behaviour. Herd behaviour is defined as any behaviour in which the probability of acceptance of an idea (or adoption of behaviour) by an individual is a factor of a group adoption. Considering a social network structure, we set a positive probability of a node in the network to accept new ideas. Although Full Herd Behaviour in social decision-making is by no doubt somewhat theoretical, we assume it to exist, to some degree as a factor which affects the acceptance of new ideas in a social network.
    For analysis purpose we simulate the change of states over time by synthetically constructed Barabasi-Albert networks. We further assume that the probability of infection follows Herd Behaviour rules. In comparison to the traditional models of infection-spread that have a fix probability of infection in a single encounter, the change of states in our model follow a different infection-spread pattern that better fits an exponential growth phenomena.
    A simulated study considered two theoretical models: Group Herd and Global Herd, in which the probability of infection is proportional to the infection rate in the group / full population, respectively. The simulations further show that in global herd model the first 1 % of the spreading is highly stochastic, while later spreading stages can be better predicted. The implications are relevant to viral marketing, spread of rumours, as well as academic publication and awareness.
    Although the research is in its preliminary stage and need to be approved against real events data, its initial results seem relevant to the modelling of various social phenomena.
  • How to Compare and Interpret Two Learnt Decision Trees From the Same Domain?

    Authors: Petra Perner (Institute of Computer Vision and Applied Computer Sciences)
    Primary area of focus / application: Mining
    Keywords: Machine learning, Decision tree induction, Interpretation of the results, Tree comparison measure
    Submitted at 8-Apr-2013 14:23 by Petra Perner
    Accepted
    16-Sep-2013 12:15 How to Compare and Interpret Two Learnt Decision Trees From the Same Domain?
    Data mining methods are widely used across many disciplines to identify patterns, rules or associations among huge volumes of data. Decision tree induction such as C4.5 is the most preferred method for classifica-tion since it works well on average regardless of the data set being used. The resulting decision tree has explana-tion capability but problems arise if the data set has been collected at different times or is enlarging and the decision tree induction process has been repeated. The resulting tree will change and the expert is questioning the trustworthy of the result. That brings us to the problem of comparing two decision trees in accordance with its explanation power. In this paper, we present a method how to compare two decision trees and how to interpret the change of the structure and the attributes in the decision tree.
  • New 'Quality' Tools

    Authors: Jonathan Smyth-Renshaw (Jonathan Smyth-Renshaw & Associates Ltd)
    Primary area of focus / application:
    Keywords: Quality tools, Strategy, Data analysis, 7 quality tools
    Submitted at 10-Apr-2013 20:14 by Jonathan Smyth-Renshaw
    Accepted (view paper)
    18-Sep-2013 10:15 Problem Session
    The goal of this extended session is to share a number of tools and techniques which I use during my work as a consultant and trainer. The principles of Six Sigma and Lean are widely taught. However, I believe there is a new 'battle' ground for business improvement philosophies. These are tools and techniques which all employees should know, understand and use in their daily work, not just a select few belts or 'other improvement rankings'.

    I will layout my proposal and use real examples and exercises to demonstrate the potential value of what I have called my 'new quality tools', my presentation is available to view for more detail.

    In short, I believe the time has come for a new Quality Renaissance and ENBIS should lead the way.
  • Bayesian Approach to the Determination of Thermophysical Properties of Materials

    Authors: Alexandre Allard (LNE), Géraldine Ebrard (LNE), Nicolas Fischer (LNE), Véronique Le Sant (LNE), Denis Rochais (CEA), Peter Harris (NPL), Clare Matthews (NPL), Louise Wright (NPL)
    Primary area of focus / application: Other: French special session
    Keywords: Thermal diffusivity, FLASH method, Bayesian framework, Metropolis-Hastings algorithm, Measurement uncertainty
    Submitted at 11-Apr-2013 10:12 by Alexandre Allard
    Accepted
    17-Sep-2013 17:50 Bayesian Approach to the Determination of Thermophysical Properties of Materials
    The determination of thermophysical properties is at the heart of modern materials characterisation. The laser flash method is used by the National Measurement Institutes for this purpose. A laser flash impacts the sample on its front face, which causes a temperature rise on the back face that is measured as a function of time to obtain the corresponding experimental thermogram. The diffusivity of the material is obtained by solving an inverse problem because it cannot be measured directly. The current state of the art in metrology for the evaluation of measurement uncertainty does not provide a reliable framework for such an evaluation in the case of an inverse problem. The diffusivity can be determined through the identification of the experimental thermogram with a theoretical one, generated from a thermal model. The evaluation of measurement uncertainty associated with such a determination may be performed but is not straightforward. We propose a methodology for the evaluation of the associated measurement uncertainty based on a Bayesian framework that takes into consideration both the measurements performed and the available knowledge of the properties of the material. Thanks to a Metropolis-Hastings algorithm, this framework consists of the estimation of two parameters that summarize the thermal exchanges in the process, and are used to to determine the diffusivity. A validation of the framework is performed on the basis of simulated data, and experimental data are considered in order to compare the results with the results obtained with current state of the art.
  • Prior Elicitation for Nonlinear Bayesian Inversion in Uncertainty Propagation

    Authors: Nicolas Bousquet (EDF R&D), Mathieu Couplet (EDF R&D), Shuai Fu (EDF R&D)
    Primary area of focus / application:
    Keywords: Bayesian statistics, Prior elicitation, Uncertainty propagation, Inverse problems, Risk and reliability, Industrial application
    Submitted at 11-Apr-2013 16:30 by Nicolas Bousquet
    Accepted
    17-Sep-2013 18:10 Prior Elicitation for Nonlinear Bayesian Inversion in Uncertainty Propagation
    We consider the problem of quantifying the uncertainty characterizing a random input of a nonlinear rule of decision, or representation functional (for instance a so-called computer code function), when realisations of this input cannot be observed. A probabilistic inversion of this nonlinear transformation is required. The case-study that motivates this study is the assessment of the distribution of the roughness coefficient of a river, which summarizes the state of the riverbed topography (and is therefore considered as stochastic). This distribution must be inferred from known discharge values and measured water levels, through a nonlinear hydraulical function. The talk will focus on the following points. First, we highlight the interest of a Bayesian statistical setting to realize such an inversion, resulting from (a) the availability of prior information, and: (b) the need for accouting for constraints arising from global sensitivity analysis. Second, based on an algorithm proposed in Fu (2012) and Fu et al. (2012), we summarize a methodology of inversion when the complex function to invert is time-consuming and illustrate it using the case-study. This methodology implies to determine sequential strategies for exploring the input space, using designs of numerical experiments, and interpolating the function, possibly using stochastic meta-modelling (e.g., kriging).

    Remaining issues in both steps are especially pointed out and discussed, as avenues for future research.


    Fu, S. (2012). Bayesian probablilistic inversion in uncertainty analysis. Ph.D. thesis, Paris-Sud (Orsay) University.

    Fu, S., Celeux, G., Bousquet, N., Couplet, M. (2012).
    Bayesian inference for inverse problems occuring in uncertainty analysis. INRIA RR-7995.