26.11.2012 Views

Technical Sessions – Monday July 11

Technical Sessions – Monday July 11

Technical Sessions – Monday July 11

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

136-701, Seoul, Korea, Republic Of, fineyouth@korea.ac.kr,<br />

Cheong Sool Park, Jun Seok Kim, Sung-Shick Kim, Jun-Geol<br />

Baek<br />

We propose a new method to improve the accuracy for multivariate classification<br />

by integrating feature selection and classification. The method is<br />

self-sufficient by using the properties of various existing algorithms to learn<br />

NN(neural network) as training information. Each learned NN reflects those<br />

properties via connecting weights. The connected NN can be learned again for<br />

classification and to produce new features. We also study algorithm combinations<br />

to determine how to use the method suitably for data characteristics. The<br />

method can be applied to various areas requiring high accuracy.<br />

4 - Cluster Validation Attitude via Information Distances<br />

Zeev (Vladimir) Volkovich, Ort Braude Academic College,<br />

Yehiam 6, 21823, Karmiel, Israel, zeev@actcom.co.il<br />

Stability of cluster solutions reveals the reliability of the clustering process.<br />

Partition’ readiness is characterized by the its clusters homogeneity represented<br />

in our approach by the information distances between samples’ occurrences in<br />

the clusters. In the framework of the Gaussian Mixture Model these distances<br />

lead to the known probability metrics such as the weighed T-square statistic.<br />

The appropriate cluster number can be deduced from the most concentrated<br />

at the origin, distance’s empirical distribution, constructed for several possible<br />

quantities of clusters.<br />

� MB-14<br />

<strong>Monday</strong>, 14:00-15:30<br />

Meeting Room 207<br />

Nonsmooth Optimization I<br />

Stream: Continuous and Non-Smooth Optimization<br />

Invited session<br />

Chair: Adil Bagirov, School of Information Technology &<br />

Mathematical Sciences, University of Ballarat, University Drive,<br />

Mount Helen, P.O. Box 663, 3353, Ballarat, Victoria, Australia,<br />

a.bagirov@ballarat.edu.au<br />

1 - Epsilon-cleaning of Big Datasets<br />

Andrew Yatsko, ITMS Dept., University of Ballarat, Mt. Helen<br />

campus, 3350, Mount Helen, VIC, Australia,<br />

andrewyatsko@students.ballarat.edu.au, Adil Bagirov<br />

In applications of Cluster Analysis and Data Classification there is a performance<br />

issue arising from the sheer multitude of instances. An existing technique<br />

does clearing of the closely packed data, but what is discarded is arbitrary.<br />

We address the issue of bias and suggest a way of preserving the cropped<br />

data. A modified procedure of cleaning is proposed and tested on some known<br />

examples. It is run recursively on reduced sets, their integrity being verified.<br />

The staged out cleaning offers an update mode from simple to complex, able to<br />

dramatically improve performance of some algorithms.<br />

2 - Lipschitzian Properties of the Dual Pair in Infinite-<br />

Dimensional Linear Optimization<br />

Marco A. López-Cerdá, Statistics and Operations Research,<br />

Alicante University, Ctra. San Vicente de Raspeig s/n, 3071,<br />

Alicante, Spain, marco.antonio@ua.es, Andrea Beatriz Ridolfi,<br />

Virginia N. Vera de Serio<br />

We apply coderivatives and other tools from variational analysis to study the<br />

stability of the feasible sets of both, the primal and the dual problem in infinitedimensional<br />

linear optimization with infinitely many explicit constraints and<br />

an additional conic constraint. After providing some specific duality results for<br />

our dual pair, we study the Lipschitz-like property of both mappings and also<br />

give bounds for the associated Lipschitz moduli.<br />

3 - Novel Robust Regression Methods Based on Nonsmooth<br />

and Derivative Free Optimization<br />

Gleb Beliakov, School of Information Tecnology, Deakin<br />

University, Melbourne, Victoria, Australia, gleb@deakin.edu.au,<br />

Andrei Kelarev<br />

Robust regression methods based on non-smooth optimization methods have<br />

many important applications. Our experiments compare the performance of<br />

several derivative free optimization algorithms computing robust multivariate<br />

estimators, where the objective is non-smooth, non-convex and expensive to<br />

calculate. It is shown that the existing algorithms often fail to deliver optimal<br />

solutions. We introduce two new methods using Powell’s derivative free<br />

algorithm. Extensive experimental results demonstrate that our new proposed<br />

methods are reliable and can process very large data sets.<br />

IFORS 20<strong>11</strong> - Melbourne MB-15<br />

4 - Algorithms for Optimization of Electricity Distribution<br />

Systems when Upgraded by Renewable Energy<br />

Sattar Seifollahi, School of Information Technology &<br />

Mathematical Sciences, University of Ballarat, Australia,<br />

s.seifollahi@ballarat.edu.au, Adil Bagirov<br />

Distributed energy planning is a non-convex combinatorial problem, and in<br />

many cases, it may contain integer variables. In this talk, we present nonsmooth<br />

optimization based algorithms for solving such problems. The algorithms<br />

can be applied to networks of large size and have advantages over<br />

the existing algorithms in that they guarantee a globally optimal solution. An<br />

overview on the use of some existing optimization algorithms in the distribution<br />

systems is also provided. The results of the proposed methods are compared<br />

with others, demonstrating the efficiency of the proposed algorithms.<br />

� MB-15<br />

<strong>Monday</strong>, 14:00-15:30<br />

Meeting Room 208<br />

Analyses for Air and Space Operations<br />

Stream: Military, Defense and Security Applications<br />

Invited session<br />

Chair: Ariela Sofer, George Mason University, MS4A6 4400<br />

University Drive, 22030, Fairfax, VA, United States, asofer@gmu.edu<br />

1 - Predicting the Required Naval Combat Helicopter Fleet<br />

Size<br />

David Marlow, Air Operations Division, DSTO, 506 Lorimer St,<br />

3207, Fishermans Bend, Vic, Australia,<br />

david.marlow@dsto.defence.gov.au, Ana Novak<br />

A discrete event simulation has been developed that models a fleet of naval<br />

combat helicopters. The simulation includes detailed models of the ashore<br />

and embarked flying program, and represents unscheduled, phased and deep<br />

maintenance. The purpose of the model is to assist the Australian Government<br />

in determining the size of the new fleet of naval combat helicopters. The<br />

fleet must meet minimum requirements for the number of helicopters embarked<br />

on ships, while simultaneously providing a minimum number of annual flying<br />

hours ashore.<br />

2 - Routing Optimisation for Air-to-Air Refuelling<br />

Yue-Jin Wang, Joint Operations Division, Defence Science and<br />

Technology Organisation, DSTO-Fairbairn, Canberra, ACT,<br />

Australia, yue-jin.wang@dsto.defence.gov.au, Ian Brunskill<br />

Air-to-Air refuelling (AAR) is the process of transferring fuel from one aircraft<br />

(tanker) to another (receiver) during flight. Optimal use of limited tanker resources<br />

is a major concern in military AAR planning. This paper formulates<br />

the tanker assignment problem as a multiple-vehicle routing problem with time<br />

windows (VRPTW). It aims at designing a set of minimum-cost routes for a<br />

tanker fleet tasked to refuel a number of receiver aircraft at predefined locations<br />

in the required time windows. A genetic algorithm-based technique is<br />

developed for solving this complicated VRPTW problem.<br />

3 - Analysis Tools in Swedish Air Force Studies<br />

Anders Tavemark, Division of Defence Analysis, Swedish<br />

Defence Research Agency, FOI, SE-172 90, Stockholm, Sweden,<br />

tavemark@foi.se<br />

Swedish operational analysts support military studies at the Swedish Armed<br />

Forces. The studies range from system studies at the armed services level to<br />

long term planning studies at the armed forces level. Different study questions<br />

require different study approaches. One constant limitation is available resources.<br />

This presentation describes experiences from using different analysis<br />

tools in Swedish Air Force studies, ranging from spreadsheets to the Swedish<br />

Air Force Combat Simulation Centre.<br />

4 - Scheduling Optimization for Multi-Satellite Constellations<br />

using Column Generation<br />

Ariela Sofer, George Mason University, MS4A6 4400 University<br />

Drive, 22030, Fairfax, VA, United States, asofer@gmu.edu<br />

17

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!