Tomas Fabian

Differential evolution dynamics analysis by complex networks

Differential evolution is a simple yet efficient heuristic originally designed for global optimization over continuous spaces that has been used in many research areas. The question how to improve its performance is still popular and during the years, many successful methods dealing with optimal setting or hybridization of the control parameters were proposed. In this paper, we propose a novel approach based on modeling of the differential evolution dynamics by complex networks. In each generation, the individuals are mapped to the nodes and the relationships between them are modeled by the edges of the graph. Thanks to this simple visualization, the interconnection between the differential evolution convergence speed and the weighted clustering coefficients has been revealed. As a consequence, we have focused on the parents selection in the mutation step where the individuals are not selected randomly as usual but on the basis of their weighted clustering coefficients. Our enhancement has been incorporated in the classical differential evolution, self-adaptive differential evolution (jDE) and differential evolution with composite trial vector generation strategies and control parameters. Finally, a set of well-known benchmark functions (including 21 functions) has been used to test and evaluate the performance of the proposed enhancement of the differential evolution. The experimental results and statistical analysis indicate that the enhanced algorithms perform better or at least comparable to their original versions and the analysis of the differential evolution dynamics with the aim of the complex network might be an effective tool to improve the differential evolution convergence in the future.

Differential Evolution Enhanced by the Closeness Centrality: Initial Study

The closeness centrality can be considered as the natural distance metric between pairs of nodes in connected graphs. This paper is the initial study of the influence of the closeness centrality of the graph built on the basis of the differential evolution dynamics to the differential evolution convergence rate. Our algorithm is based on the principle that the differential evolution creates graph for each generation, where nodes represent the individuals and edges the relationships between them. For each individual the closeness centrality is computed and on the basis of its value the individuals are selected in the mutation step of the algorithm. The higher value of the closeness centrality means the higher probability to become the parent in the mutation step. This enhancement has been incorporated in the classical differential evolution and a set of 21 well-known benchmark functions has been used to test and evaluate the performance of the proposed enhancement of the differential evolution. The experimental results and statistical analysis indicate that the enhanced algorithm performs better or at least comparable to its original version.

Parking Lot Occupancy Detection Using Computational Fluid Dynamics

In this paper, we present an algorithm for estimating the occupancy of individual parking spaces. Our method is based on a computer analysis of images obtained by a camera system monitoring the activities on a parking lot. The main idea is to use the a priori available information about the parking lot geometry and the general shape of common cars to obtain a reliable status of a parking space. We strive to avoid the training phase as much as possible to reduce the time required to bring the system into a fully operational state. To achieve this goal, we focus on a probabilistic car model and a physically based feature extraction using computational fluid dynamics. Despite the fact that the very first system of a similar type has appeared more than forty years earlier, this area is still an active research topic and a completely satisfactory solution has not been found yet.

Mixture of Gaussians exploiting histograms of oriented gradients for background subtraction

Visual surveillance systems include a wide range of related areas ranging from motion detection, moving object classification and tracking to activity understanding. Typical applications include traffic surveillance, CCTV security systems, road sign detection. Each of the above-mentioned applications relies greatly on proper motion segmentation method. Many background subtraction algorithms have been proposed. Simple yet robust frame differencing, statistically based Mixture of Gaussians or sophisticated methods based on wavelets or the optical flow computed by the finite element method. In this paper we focus on novel modification of well known MoG. The intrinsic motivation stems from the inability of regular MoG implementation to handle many camera related phenomena. Here presented method exploits Histograms of Oriented Gradients to significantly reduce the influence of camera jitter, automatic iris adjustment or exposure control causing severe degradation of foreground mask. The robustness of introduced method is shown on series of video sequences exhibiting mentioned phenomena.

Differential Evolution Dynamics Modeled by Longitudinal Social Network

Abstract Differential evolution (DE) is a population-based algorithm using Darwinian and Mendel principles to find out an optimal solution to difficult problems. In this work, the dynamics of the DE algorithm are modeled by using a longitudinal social network. Because a population of the DE algorithm is improved in generations, each generation of DE algorithm is represented by one short-interval network. Each short-interval network is created by individuals contributing to population improvement. On the basis of this model, a new parent selection in the mutation operation is presented and a well-known benchmark set CEC 2013 Special Session on Real-Parameter Optimization (including 28 functions) is used to evaluate the performance of the proposed algorithm.

Small-world hidden in differential evolution

Differential evolution is an effective population-based global optimizer which is used in many areas of research. The population consists of individuals, which are mutated, crossed and better of them survive to the next generation. In this paper, we look at this process as at the communication between individuals which can be modeled by the network where the individuals are represented by the nodes and the edges between them reflect the dynamics in the population, i.e. interactions between individuals. The main goal of this work is to find out if the differential evolution algorithm is able to create the networks where the small-world phenomenon (known as six degrees of separation) is observed. The secondary objective was to investigate the dependency between the type of the selected test function and the extent of this phenomenon. To evaluate the performance of the algorithm eleven test functions from the benchmark set CEC 2015 have been used. The analysis of the generated networks indicates that the differential evolution is able to create small-world networks in majority of test functions. As the result, the selected test functions can be classified into three categories which binds to the degree of cooperation between the individuals in the population.

An algorithm for iris extraction

In this paper, we describe a new method for detecting iris in digital images. Our method is simple yet effective. It is based on statistical point of view when searching for limbic boundary and rather analytical approach when detecting pupillary boundary. It can be described in three simple steps; firstly, the bright point inside the pupil is detected; secondly, outer limbic boundary is found via statistical measurements of outer boundary points; and thirdly, inner boundary points are found by means of defined cost function maximization. Performance of the presented method is evaluated on series of iris close-up images and compared with the traditional Hough method as well.

Analysis of causality-driven changes of diffusion speed in non-Markovian temporal networks generated on the basis of differential evolution dynamics

Abstract Differential evolution (DE) is one popular meta-heuristic, which is used to solve difficult optimization problems. In the last years, a huge number of new variants of the differential evolution has been introduced to outperform previously presented algorithms. To provide solutions of higher quality or to speed-up the convergence principles as control parameters adaptation, novel mutation strategies, or combination of different mutation strategies are often used. In this work, five different variants of the differential evolution have been chosen with the goal to investigate their inner dynamics, especially spread of positive genomes within the population. To capture relationships between individuals, temporal networks, more precisely contact sequences, are used. Based on the empirical results, we have concluded that temporal networks generated on the basis of the DE algorithms dynamics are non-Markovian temporal networks. For this reason, to analyze the causality-driven changes of diffusion speed in these networks, analytical methods described by Scholtes et al. have been used.

Differential evolution based on node strength

In this paper, three novel algorithms for optimisation based on the differential evolution algorithm are devised. The main idea behind those algorithms stems from the observation that differential evolution dynamics can be modelled via complex networks. In our approach, the individuals of the population are modelled by the nodes and the relationships between them by the directed lines of the graph. Subsequent analysis of non-trivial topological features further influence the process of parent selection in the mutation step and replace the traditional approach which is not reflecting the complex relationships between individuals in the population during evolution. This approach represents a general framework which can be applied to various kinds of differential evolution algorithms. We have incorporated this framework with the three well-performing variants of differential evolution algorithms to demonstrate the effectiveness of our contribution with respect to the convergence rate. Two well-known benchmark sets (including 49 functions) are used to evaluate the performance of the proposed algorithms. Experimental results and statistical analysis indicate that the enhanced algorithms perform better or at least comparable to their original versions.

A Vision-Based Algorithm for Parking Lot Utilization Evaluation Using Conditional Random Fields

In this paper, we present an algorithm for estimating the occupancy of individual parking spaces. Our method is based on a computer analysis of images obtained by a camera system monitoring the activities on a parking lot. The proposed method extensively uses a priori information about the parking lot layout and the general shape of well-parked cars, which is incorporated in a simplified probabilistic car model. Discriminative features are extracted from a normalized image of every parking space, the relevance of these gradient-based features is prioritized via a selective flow, and furthermore, their spatial relationship is revealed through an undirected graphical model. We strive to avoid the training phase to reduce the time required to bring the system into a fully operational state. The reliability of the here devised approach is evaluated on the set of video sequences captured during different phases of a day and the results are compared against the ground truth data.