|
|
Automated Multi-Objective Parallel Evolutionary Circuit Design and Approximation
Hrbáček, Radek ; Fišer, Petr (oponent) ; Trefzer,, Martin (oponent) ; Sekanina, Lukáš (vedoucí práce)
Recently, energy efficiency has become one of the most important properties of computing platforms, especially because of limited power supply capacity of battery-power devices and very high consumption of growing data centers and cloud infrastructure. At the same time, in an increasing number of applications users are able to tolerate inaccurate or incorrect computations to a certain extent due to the imperfections of human senses, statistical nature of data processing, noisy input data etc. Approximate computing, an emerging paradigm in computer engineering, takes advantage of relaxed functionality requirements to make computer systems more efficient in terms of energy consumption, computing performance or complexity. Error resilient applications can achieve significant savings while still serving their purpose with the same or a slightly degraded quality. Even though new design methods for approximate computing are emerging, there is a lack of methods for automated approximate HW/SW design offering a rich set of compromise solutions. Conventional methods often produce solutions that are far from an optimum. Evolutionary algorithms have been shown to bring innovative solutions to complex design and optimization problems. However, these methods suffer from several problems, such as the scalability or a high number of fitness evaluations needed to evolve competitive results. Finally, existing methods are usually single-objective whilst multi-objective approach is more suitable in the case of approximate computing. In this thesis, a new automated multi-objective parallel evolutionary algorithm for circuit design and approximation is proposed. The method is based on Cartesian Genetic Programming. In order to improve the scalability of the algorithm, a brand new highly parallel implementation was proposed. The principles of the NSGA-II algorithm were used to provide the multi-objective design and approximation capability. The performance of the implementation was evaluated in multiple different applications, in particular (approximate) combinational arithmetic circuits design, bent Boolean functions discovery and approximate logic circuits for TMR schema. In these cases, important improvements with respect to the state of the art were obtained.
|
|
|
Multi-objective genetic algorithms in road traffic prediction
Petrlík, Jiří ; Brandejský, Tomáš (oponent) ; Snášel,, Václav (oponent) ; Sekanina, Lukáš (vedoucí práce)
The understanding of the road traffic behavior is a key to effective traffic control, management and organization. This task is becoming more and more important with increasing traffic demands and the number of registered vehicles. The information about the current and future traffic situation is very important for drivers and traffic operators. Fortunately, there was a huge progress in technologies for traffic data acquisition in the last few decades. Stationary sensors, such as loop detectors, radars, cameras and infrared sensors can be installed on important locations of the roads and measure various microscopic and macroscopic traffic variables. However, some measurements can lead to an incorrect data which cannot further be used in the subsequent processing tasks such as traffic prediction or intelligent control. For example, this can be caused by equipment failures or data transmission problems. It is highly desirable to have a framework, which is capable of estimating the missing values in traffic data. It is also very important to provide a reliable short-time prediction of the traffic state. In this thesis, we focus on selected problems from this domain - the imputation of missing traffic data, short time traffic forecasting and travel times estimation. The proposed solution is based on combining the state-of-the art machine learning methods such as support vector regression (SVR) with the multi-objective evolutionary optimization. SVR has various meta-parameters which should be properly set in order to achieve the best performance. The performance also strongly depends on the selection of the input variables for SVR. We used the multi-objective optimization to find the proper settings of SVR meta-parameters and input variables. Using the multi-objective optimization, we obtained many different non-dominated solutions from Pareto front. These solutions can dynamically be switched according to the traffic data which are currently available, in order to maximize the quality of prediction. The proposed methods are specially designed for environments with many missing values in traffic data. We evaluated the proposed methods using real world data and compared them with the state of the art methods for the traffic data imputation and short term prediction such as the probabilistic principal component analysis and support vector regression optimized by a single objective optimization. The proposed methods provide better results than these state of the art methods especially in the cases where there are many missing values in the traffic data.
|
|
|
Multiobjective optimization of electromagnetic structures based on self-organizing migration
Kadlec, Petr ; Prof. Hans L. Hartnagel (oponent) ; Škvor,, Zbyněk (oponent) ; Raida, Zbyněk (vedoucí práce)
This thesis describes a novel stochastic multi-objective optimization algorithm called MOSOMA (Multi-Objective Self-Organizing Migrating Algorithm). It is shown that MOSOMA is able to solve various types of multi-objective optimization problems (with any number of objectives, unconstrained or constrained problems, with continuous or discrete decision space). The efficiency of MOSOMA is compared with other commonly used optimization techniques on a large suite of test problems. The new procedure based on finding of minimum spanning tree for computing the spread metric for problems with more than two objectives is proposed. Recommended values of parameters controlling the run of MOSOMA are derived according to their sensitivity analysis. The ability of MOSOMA to solve real-life problems from electromagnetics is shown in a few examples (Yagi-Uda and dielectric filters design, adaptive beam forming in time domain…).
|
|
|
Modul plánování a rozlosování soutěží
Jelínek, Zdeněk ; Soukup, Ondřej (oponent) ; Křivka, Zbyněk (vedoucí práce)
Tato bakalářská práce popisuje návrh a implementaci modulu k rozlosování turnajů v kompetitivních sportech. Modul páruje hráče tak, aby bylo minimalizováno opakování zápasů, páry soupeřů hrály co nejkvalitnější zápasy, a zároveň, aby si každý hráč zahrál příslušný zápasů, a to s polynomiální nejhorší asymptotickou časovou složitostí.
|
|
|
Pokročilé optimalizační modely v oblasti oběhového hospodářství
Pluskal, Jaroslav ; Bednář, Josef (oponent) ; Šomplák, Radovan (vedoucí práce)
Diplomová práce se zabývá aplikací optimalizačních metod v oblasti oběhového hospodářství. Úvod je zaměřen na vysvětlení hlavních bodů této problematiky a její přínosy pro ekonomiku a životní prostředí. Dále jsou uvedeny překážky bránící v přechodu ze současného nakládání s odpady. V práci je popsán matematický aparát, který je dále využit v praktické části. Jádrem práce je matematický optimalizační model implementovaný v~softwaru GAMS a generátor vstupních dat zpracovaný ve VBA. Model zahrnuje všechny významné způsoby nakládání s odpady s ohledem na ekonomické i ekologické aspekty včetně dopravy. Funkčnost je následně předvedena na malé úloze. Stěžejním výsledkem práce je aplikace modelu na reálných datech týkajících se ČR. Na závěr je provedena analýza výpočtové náročnosti vzhledem k rozsahu úlohy.
|
|
|
Antenna Arrays with Synthesized Frequency Response of Gain
Všetula, Petr ; Polívka,, MIlan (oponent) ; Bonefačic, Davor (oponent) ; Raida, Zbyněk (vedoucí práce)
In the thesis, we present a method of the synthesis of a dipole antenna array with prescribed spectral and spatial filtering capabilities. Thanks to the spatial filtering capabilities, the main lobe direction and the value of gain vary negligibly over the operating band. Thanks to the spectral filtering capabilities, the value of gain is maximal in the operating band and minimal out of the operating band. In order to synthesize a dipole array with prescribed filtering capabilities, amplitudes, phases and dimensions of antenna elements are optimized. The initial optimization is speeded up by considering an idealized antenna array when evaluating objective functions. Since the optimization comprises requirements on the main lobe direction, the value of gain and impedance matching, a multi-objective optimization is used. The optimized antenna array is analyzed by a full-wave simulator to verify results of the synthesis. Finally, the synthesized dipole array is manufactured and its performance is experimentally verified.
|
|
|
Multi-Objective Optimization of EM Structures With Variable Number of Dimensions
Marek, Martin ; Vrba,, Jan (oponent) ; Šenkeřík,, Roman (oponent) ; Kadlec, Petr (vedoucí práce)
This dissertation thesis deals with multi-objective evolutionary optimization algorithms with a variable number of dimensions. Such an algorithm enables us to solve optimization tasks that are otherwise solved only by assuming unnatural simplifications. The research of the optimization algorithms with a variable number of dimensions required the development of a new optimization framework. This framework contains, apart from various optimization methods including two novel multi-objective algorithms for a variable number of dimensions -- VND-GDE3 and VND-MOPSO, a library of various benchmark problems. A set of multi-objective benchmark problems with a variable number of dimensions is a part of the library designed to assess and verify the novel methods with a variable number of dimensions. Novel methods are exploited on several miscellaneous real-life optimization problems in the final chapter of this thesis.
|
|
|
Toolbox pro vícekriteriální optimalizační problémy
Marek, Martin ; Hurák,, Zdeněk (oponent) ; Kadlec, Petr (vedoucí práce)
Tato práce se zabývá problematikou více-kriteriálních optimalizací. Je vysvětleno, která řešení jsou optimální při použití více konfliktních kriteriálních funkcí a jak tato optimální řešení (Paretovo čelo) v množině možných řešení vyhledat. Poté jsou popsány principy algoritmů NSGA-II, MOPSO a GDE3. V následujících kapitolách jsou představeny testovací metriky a problémy. Na závěr práce jsou tyto tři algoritmy porovnány na základě několika metrik.
|
|
|
Evolutionary Design of EEG Data Classifier
Kuželová, Simona ; Jawed, Soyiba (oponent) ; Mrázek, Vojtěch (vedoucí práce)
This thesis focuses on developing an effective classifier for candidate classification based on a set of extracted Electroencephalography (EEG) signal features. To achieve this, a genetic algorithm was utilized for feature selection and optimalization of the classifier’s parameters based on five criteria: minimizing the number of features, minimizing inference time, and maximizing classification sensitivity, specificity, and accuracy. The eyes opened EEG data of 31 candidates suffering from Major Depressive Disorder (MDD) and 28 healthy candidates were used for feature extraction, with the goal of classifying candidates as either having MDD or being healthy. Two algorithms, NSGA-II and NSGA-III, were tested. The proposed algorithm operated with three criteria, but two additional criteria, sensitivity and specificity, were added. NSGA-III was more effective in this case and was used in the remaining experiments. Constraints were introduced to improve performance, and different values for the mutation and crossover probability were tried. The classifiers from the final result have an average accuracy of $91.36\%$, sensitivity of $91.82\%$, and specificity of $90.84\%$. In the final experiments most frequently used channels were F3 and C3 channels and most commonly utilized waveband was gamma waveband. Overall, this work presents effective classifiers that were obtained using the proposed algorithm, which utilizes a genetic algorithm for parameter optimization.
|
|
|
Multi-Objective Optimization of Complex Composite Structures with Variable Stiffness
Symonov, Volodymyr ; Halama, Radim (oponent) ; Píštěk, Antonín (oponent) ; Juračka, Jaroslav (vedoucí práce)
The thesis is dedicated to development of a multi-objective optimization methodology for complex composite structures with variable stiffness. A multi-level hybrid optimization algorithm is developed based on a hybrid optimization method with interpolating response surface, a Genetic Algorithm and a one-dimensional optimization. Finite element analysis software MSC Nastran is used for structural analyses. A new Genetic Algorithm and a parallel one-dimensional optimization algorithm based on “Golden section” method are developed for the methodology. The finite element analysis software and the developed optimization algorithms are integrated with help of a commercial optimization software Noesis Optimus by Noesis Solutions. The developed methodology is verified on an example optimization problem. The results of the problem optimization are compared to those obtained using previously developed methodologies.
|
host ::
RSS.