Jarno Martikainen

A general framework for statistical performance comparison of evolutionary computation algorithms

This paper proposes a statistical methodology for comparing the performance of evolutionary computation algorithms. A twofold sampling scheme for collecting performance data is introduced, and these data are analyzed using bootstrap-based multiple hypothesis testing procedures. The proposed method is sufficiently flexible to allow the researcher to choose how performance is measured, does not rely upon distributional assumptions, and can be extended to analyze many other randomized numeric optimization routines. As a result, this approach offers a convenient, flexible, and reliable technique for comparing algorithms in a wide variety of applications.

Fitness Function Approximation by Neural Networks in the Optimization of MGP-FIR Filters

In this paper we introduce a neural network based method for speeding up the fitness function calculations in a genetic algorithm (GA)-driven optimization process of multiplicative general parameter finite impulse response (MGP-FIR) filters. In this case, calculating the fitness of a candidate solution is an extensive and time-consuming task. However, our results show that it is possible to approximate the fitness function components with neural networks up to sufficient degree, thus enabling the genetic algorithm to perform the fitness calculations considerably faster. This allows the algorithm to evaluate larger number of generations in a given time. Our results suggest that it is possible to decrease the approximation error of the neural network so that the NN-assisted GA eventually offers competitive performance compared to a reference GA

Time-dependent performance comparison of evolutionary algorithms

We propose a statistical methodology for comparing the performance of evolutionary algorithms that iteratively generate candidate optima over the course of many generations. Performance data are analyzed using multiple hypothesis testing to compare competing algorithms. Such comparisons may be drawn for general performance metrics of any iterative evolutionary algorithm with any data distribution. We also propose a data reduction technique to reduce computational costs.

Comparison of a Fuzzy EP Algorithm and an AIS in Dynamic Optimization Tasks

In this paper we compare a specific evolutionary programming algorithm with a basic artificial immune system-based method in a dynamic combinatorial optimization task. Evolutionary algorithms are known to produce competitive results in optimization tasks, where only a single best solution is desirable. Artificial immune systems, however, can simultaneously find many different competitive solutions, and this property makes them an interesting choice in dynamic optimization environments. The performance of these two algorithms is compared using a nonparametric statistical framework that does not require any knowledge regarding the output distribution of the algorithms

Designing Multiplicative General Parameter Filters Using Adaptive Genetic Algorithms

Multiplicative general parameter (MGP) approach to finite impulse response (FIR) filtering introduces a novel way to realize cost effective adaptive filters in compact very large scale integrated circuit (VLSI) implementations used for example in mobile devices. MGP-filter structure comprises of additions and only a small number of multiplications, thus making the structure very simple. Only a couple of papers have been published on this recent innovation and, moreover, MGP-filters have never been designed using adaptive genetic algorithms (GA). The notion suggesting the use of adaptive parameters is that optimal parameters of an algorithm may change during the optimization process, and thus it is difficult to define parameters beforehand that would produce competitive solutions. In this paper, we present results of designing MGP-FIR basis filters using different types of adaptive genetic algorithms, and compare the results to the ones obtained using a simple GA.

Organizing an online soft computing conference: a case study

The Internet provides ever increasing ways to communicate and spread information. The Institute of Intelligent Power Electronics at the Helsinki University of Technology organized an online conference during fall 2000. This new type of conference proved to be a success, but a lot more could be achieved. We sum up the experience gathered from organizing WSC5.

Promoting polynomial predictive filtering on the Internet

Polynomial predictive filtering (PPF) is a powerful tool for modern digital signal processing. Still, not as widely known as it should be. To make the new techniques easier to approach, we have established an Internet site, which, at the moment contains illustrative documentation for the recursive linear smoothed Newton (RLSN) and the Heinonen-Neuvo (H-N) finite impulse response (FIR) predictors. Two MATLAB-based easy-to-use filter designers for both the RLSN and the H-N predictors are also available. With the help of these automatic designers, users are offered a convenient way to get to know what polynomial predictive filtering is all about by easily experimenting themselves. The Internet, and the World-Wide Web (WWW) especially, offer a flexible, easily maintainable and popular platform for promoting these ideas and techniques.

Optimizing Multiplicative General Parameter Finite Impulse Response Filters Using Evolutionary Computation

Summary. Evolutionary algorithms have been studied in the context of optimization problems for decades. These nature-inspired methods have lately received increasing amount of attention especially among demanding practical optimization tasks. This Chapter describes how evolutionary algorithms can be used to increase the speed and robustness of the design process of a digital filter used in power systems instrumentation. Designing such filters is a challenging optimization problem due to a discrete and exhaustive search space and time-consuming evaluation of the objective function. This chapter introduces new approaches to evolutionary computation using adaptive parameters, hierarchical populations, and the fusion of neural networks and evolutionary algorithms to tackle the bottlenecks of this challenging design process.

Accelerating optimization using probabilistic affinity evaluation and Clonal Selection Principle

The performance of evolutionary algorithms in optimization is tightly coupled to the computational effort required by the evaluation of the objective function. If the objective function is too expensive to evaluate, then, the elaboration of the procedures of the search algorithm alone may not result in the required improvement in algorithms performance. However, if there is a way to speed up or decrease the number of objective function evaluations, even a basic algorithms can potentially achieve better results due to the increased number of generation run in given time. This paper considers a probabilistic objective function evaluation scheme in which the candidate solutions are evaluated and evolved based on their objective function value.