Session I (Chair: Aleš Zamuda) 9:00 – 10:30

1. Analysis of Two Algorithms for Multi-Objective Min-Max Optimization

   S. Alicino, M. Vasile

   This paper presents two memetic algorithms to solve multi-objective min-max problems, such as the ones that arise in evidence-based robust optimization. Indeed, the solutions that minimize the design budgets are robust under epistemic uncertainty if they maximize the Belief in the realization of the value of the design budgets. Thus robust solutions are found by minimizing with respect to the design variables the global maximum with respect to the uncertain variables. A number of problems, composed of functions whose uncertain space is modelled by means of Evidence Theory, and presenting multiple local maxima as well as concave, convex, and disconnected fronts, are used to test the performance of the proposed algorithms. [paper, presentation]

2. Comparison Between Single and Multi Objective Genetic Algorithm Approach for Optimal Stock Portfolio Selection

   N. Cvörnjek, M. Brezočnik, T. Jagrič, G. Papa

   Portfolio selection is one of the most common problem in the field of finance. Many investors would like to allocate their funds in such way that ratio between return and risk will be as high as possible. Up to today, the problem has been solved with various approaches based on genetic algorithm technique and GA has proved to be suitable. In this paper we applied two different approaches based on genetic algorithm technique in order to solve the problem. First is single objective approach and second is multi objective one (NSGA-II). Results are showing that there is no significant difference between approaches. [paper, presentation]

3. Simulation-Based GA Optimization for Production Planning

   J. E. Diaz Leiva, J. Handl

   Effective production planning requires models that are capable of accounting for the complexity and uncertainty intrinsic to manufacturing systems. While the identification of a globally optimal plan is desirable, a more important requirement is the ability of a model to produce production plans that are sufficiently realistic to be implemented in practice and are robust to perturbations in the system. Here, we present a simulation-based optimization approach that employs discrete event simulation and a genetic algorithm as a methodology to support decision making in the area of production planning. The model aims to minimize the sum of expected backorders and inventory costs, while incorporating system constraints and the uncertainty that derives from variations of manufacturing lead times, occurrence of work centre failures and repair service times. Preliminary results for a real-world problem indicate that the model is capable of producing feasible production plans that correctly account for the uncertainty intrinsic to the underlying manufacturing system. [paper, presentation]

Session II (Chair: Peter Korošec) 11:00 – 12:30

1. Multi-Population Adaptive Inflationary Differential Evolution

   M. Di Carlo, M. Vasile, E. Minisci

   In this paper, a multi-population version of Adaptive Inflationary Differential Evolution, which automatically adapts the crossover probability and the differential weight of the Differential Evolution, is presented. The multi-population algorithm exploits the use of different populations, and the local minima found by each population, to assess the distance between minima; a probabilistic kernel based approach is then used to automatically adapt the dimension of a bubble in which the population is re-initialized after converging to a local minimum. The algorithm is tested on two real case functions and on two difficult academic functions. [paper, presentation]

2. Automated Slogan Production Using a Genetic Algorithm

   P. Tomašič, G. Papa, M. Žnidaršič

   Invention of slogans is an intelligent and highly creative task. As such, it is a challenging problem for computational methods. In this paper we present our solution based on the use of linguistic resources and evolutionary computing. [paper, presentation]

3. A Comparison of Search Spaces and Evolutionary Operators in Facial Composite Construction

   J. J. Mist, S. J. Gibson, C. J. Solomon

   In this paper three experiments concerning the use of interactive evolutionary algorithms in the creation of facial composites are reported. A reduced dimension human evaluation based search space is created from a larger search space using a pairwise face comparison task. The human reduced search space is used in the comparison of two mutation operators and two recombination operators. Finally, three search spaces are compared: large, human reduced, and a mathematically reduced search space. No statistically significant differences are found between the performances of the operators or the search spaces. [paper, presentation]

Session III (Chair: Gregor Papa) 14:00 – 15:30

1. Local Search Based Optimization of a Spatial Light Distribution Model

   D. Kaljun, J. Žerovnik

   Recent development of LED technology enabled production of lighting systems with nearly arbitrary light distributions. A nontrivial engineering task is to design a lighting system or a combination of luminaries for a given target light distribution. Here we use heuristics for solving the problem restricted to symmetrical distributions. A genetic algorithm and several versions of local search heuristics are compared showing that practically useful approximations can be achieved with majority of the algorithms. [paper, presentation]

2. Parallel CUDA Implementation of the Desirability-Based Scalarization Approach for Multi-Objective Optimization Problems

   E. Akça, Ö. T. Altinöz, S. U. Emel, A. E. Yilmaz, M. Efe, T. Yaylagul

   In this study, we present the results obtained for the parallel CUDA implementation of the previously proposed desirability-based scalarization approach for the solution of the multi-objective optimization problems. Our simulations show that compared to the sequential Java implementation, it is possible to find the same solutions (up to 16-time faster manner) by parallel CUDA implementation. We also try to outline our experiences of troubleshooting throughout the implementation as guidelines for upcoming researchers working in the same field. [paper, presentation]

3. Differential Evolution for Self-Adaptive Triangular Brushstrokes

   U. Mlakar, J. Brest, A. Zamuda

   This paper proposes a lossy image representation where a reference image is approximated by an evolved image, constituted of variable number of triangular brushstrokes. The parameters of each triangle brush are evolved using differential evolution, which self-adapts the triangles to the reference image, and also self-adapts some of the control parameters of the optimization algorithm, including the number of triangles. Experimental results show the viability of the proposed encoding and optimization results on a few sample reference images. [paper, presentation]

Session IV (Chair: Jurij Šilc) 16:00 – 17:30

1. Extended Finite-State Machine Inference with Parallel Ant Colony Based Algorithms

   D. Chivilikhin, V. Ulyantsev, A. Shalyto

   This paper addresses the problem of inferring extended finite-state machines (EFSMs) with parallel algorithms. We propose a number of parallel versions of a recent EFSM inference algorithm MuACO. Two of the proposed algorithms demonstrate super-linear speedup. [paper, presentation]

2. Empirical Convergence Analysis of Genetic Algorithm for Solving Unit Commitment Problem

   D. Butala, D. Velušček, G. Papa

   This paper presents an implementation and empirical convergence analysis results of genetic algorithm for solving unit commitment problem in a power market. Various parameter settings are presented including an algorithm with a sequence of parameters, also called a variable-structure genetic algorithm. Implemented algorithm successfully solves both small and large scale problems and shows how much more efficient variable-structure genetic algorithm is in practice. [paper, presentation]

3. General Discussion

   P. Korošec, G. Papa, J. Šilc, and A. Zamuda