1,720,964 research outputs found
A hybrid of Bayesian-based global search with Hooke–Jeeves local refinement for multi-objective optimization problems /
Get PDFThe proposed multi-objective optimization algorithm hybridizes random global search with a local refinement algorithm. The global search algorithm mimics the Bayesian multi-objective optimization algorithm. The site of current computation of the objective functions by the proposed algorithm is selected by randomized simulation of the bi-objective selection by the Bayesian-based algorithm. The advantage of the new algorithm is that it avoids the inner complexity of Bayesian algorithms. A version of the Hooke–Jeeves algorithm is adapted for the local refinement of the approximation of the Pareto front. The developed hybrid algorithm is tested under conditions previously applied to test other Bayesian algorithms so that performance could be compared. Other experiments were performed to assess the efficiency of the proposed algorithm under conditions where the previous versions of Bayesian algorithms were not appropriate because of the number of objectives and/or dimensionality of the decision space
Solving bicriteric optimisation problem of traveling salesman using genetic algorithms.
No full textMy thesis is about biobjective Travel Salesman Problem (TSP) which was solved by genetic algorithm. My goal was to analyse the hyperparameters of genetic algorithm and get which values best solve this problem. To make sure if I am getting efficient results I compared my implemented genetic algorithm results with Branch and Bound algorithm which basically solves biobjective TSP and gets optimal Pareto front of 17 cities. During analysis of hyperparameters, I took into account not only the performance of genetic algorithm but also the running time of an algorithm because in reality time is very crucial when you are a courier and need an optimal multi-city route right away. Described and implemented genetic algorithm managed to find the closest results to the optimal solution in a short period of time
Game controller training with different machine learning methods.
No full textMachine learning has existed for decades, and major early works by Arthur Samuel laid the foundations for the field. Larger datasets and more powerful computers, which have become available over the past 10 years, have significantly increased the potential of machine learning in a variety of fields, such as medicine, engineering, biology, and machine learning has also improved in computer games. Analyzing the data in these areas solves such tasks as classification, clustering, optimization, forecasting. Often, these types of tasks are difficult to solve with traditional methods and algorithms, so to solve these problems quickly and efficiently, different machine learning methods that operate on the basis of artificial intelligence are used to help. Numerous videos can be found on the popular website YouTube, where the computer is trained to play various games using machine learning methods. Most of the demonstrated machine learning methods achieve excellent results, but objective comparisons between them are quite rare. Both algorithms analyzed in this work are widely applied in various fields, they are also suitable for solving game learning problems. One of them, genetic algorithms and neural networks, is a particularly powerful tool that is widely used not only for games, but also for solving problems of various scales and spectra. This work demonstrates the advantage of specific genetic algorithm operations: crossover, selection, and mutations in the controller learning process and their successful application to controller evolution. On the other hand, the reinforcementlearning algorithm is also widely used in the market (used in the financial sector to analyze exchange strategies, in the energy sector to assess the security of power systems, etc.). The reinforcement learning algorithm explores how agents can consistently perform actions in a given environment to maximize long-term reward (this is exactly the feature illustrated by the work done). Reinforcement training also uses a fairly well-understood and mathematically based system of Markov decision processes, while genetic algorithms are largely based on heuristics. Based on the results of this experiment, it can be concluded that the neural network with a hidden layer, trained by genetic algorithms, proved to be better than the agent trained by reinforcement learning methods. A multilayer neural network can achieve impressive results faster than an agent trained in an reinforcement learning approach
Efficiency improvement analysis of mask r-cnn.
No full textNeural networks are a large number of interconnected mathematical neural models. Neural networks are very widely used to solve a variety of problems: speech synthesis and recognition problems, computer vision problems, and others. Convolutional neural networks can be applied to solve problems of object localization and segmentation. Solutions to these problems are widely applied in the following areas: autonomous car safety (accurate real-time detection of pedestrians, obstacles, road signs), mechanics and robotics (machine monitoring and control, real-time identification of defective products, etc.), disease detection (cancer cell detection in photos, etc.), transportation (route planning, finding the shortest route, etc.) and others. The Mask-RCNN network is used to address these issues. Mask R-CNN is a deep convolutional neural network used to solve the problem of instance segmentation. Neural networks have many parameters, some of which are parameters that the network learns during learning, which are called model parameters, and others that do not change during learning, called hyperparameters. Certain combinations of hyperparameters allow for better network estimates in less time. Optimization algorithms can be used to discover these combinations. Several hyperparameter optimization algorithms that can be used are: manual search, random search, grid search, Bayesian optimization algorithm. The Bayesian optimization used tries to discover the global optimum in the smallest number of steps, so it is used to discover the optimal combinations of hyperparameters in the Mask R-CNN network to get the highest possible network estimate
Finding optimal rules for stock trading using genetic algorithms.
No full textThe goal of this bachelor’s thesis was to use genetic algorithms with neural networks to optimize existing trading strategies and find new profitable ways to trade securities. To achieve this goal two algorithms, used for imitating trading activities, were implemented. First one was based on genetic algorithm structure and was capable of trading using a single analytical index. Second one, based on neural networks structure, which is trained by genetic algorithm, was capable of trading using multiple analytical indices. Different analytical indices on these algorithms were analyzed, trying to determine the most profitable way to use them. Also, during the testing an additional parameter was introduced, to try and improve risk management
Neural network driven two dimention car models trained by genetic algorithms.
No full textThe purpose of bachelor‘s thesis was to train neural networks to control car models in two dimension environment, so that it could finish a lap in an unfamiliar track. Two one-criteria algorithms: genetic and backpropagation, and one two-criteria genetic algorithm were created. Experiments were conducted with all algorithms changing the number of neurons in the hidden layer of the network. All results were described, while giving possible reasons for the behaviour of the algorithms
Computational Modelling, Signal Analysis and Optimisation of Biosensors.
No full textBiosensor is a device for detection and measurement of substrate concentration in analysed solution. It is a cheap and reliable device used in environment monitoring, food industry and medicine. In dissertation an artificial neural networks is applied to the biosensor to determine multiple substrate concentrations. The case of the multiple substrate is analysed. The biosensor model involves substrate interaction and the external (Nerst) diffusion layer. The response of the biosensor is used to determine multiple substrates concentrations. Moreover, the effect of the external diffusion layer is investigated. Furthermore, the determination of concentration of two substrates using two biosensor static currents is investigated and the effect of a diffusion module is analysed. In order to make good biosensor, the device should meet many contradictory requirements. Dissertation presents a method that combines mathematical modeling, multi-objective optimisation and multi-dimensional visualisation that is intended for the design and optimisation of amperometric biosensors. The approach for optimising parameters of biosensors are based on the availability of a mathematical model of a catalytic biosensor. A multi-objective visualisation of trade-off solutions and Pareto optimal decisions is applied to select of the most favorable decision by a human expert when designing biosensors. The multi-objective optimisation was applied to glucose and phenol biosensors
A hybrid of Bayesian-based global search with hooke-jeeves local refinement for multi-objective optimization problems /
No full textBiojutiklių kompiuterinis modeliavimas, signalo analizė ir optimizavimas.
No full textBiosensor is a device for detection and measurement of substrate concentration in analysed solution. It is a cheap and reliable device used in environment monitoring, food industry and medicine. In dissertation an artificial neural networks is applied to the biosensor to determine multiple substrate concentrations. The case of the multiple substrate is analysed. The biosensor model involves substrate interaction and the external (Nerst) diffusion layer. The response of the biosensor is used to determine multiple substrates concentrations. Moreover, the effect of the external diffusion layer is investigated. Furthermore, the determination of concentration of two substrates using two biosensor static currents is investigated and the effect of a diffusion module is analysed. In order to make good biosensor, the device should meet many contradictory requirements. Dissertation presents a method that combines mathematical modeling, multi-objective optimisation and multi-dimensional visualisation that is intended for the design and optimisation of amperometric biosensors. The approach for optimising parameters of biosensors are based on the availability of a mathematical model of a catalytic biosensor. A multi-objective visualisation of trade-off solutions and Pareto optimal decisions is applied to select of the most favorable decision by a human expert when designing biosensors. The multi-objective optimisation was applied to glucose and phenol biosensors
Medžiagų koncentracijų tirpalų mišiniuose nustatymas, taikant biojutiklius ir dirbtinius neuroninius tinklus
Get PDFBiosensor response, in case of multi-substrate mixture, has nonlinear dependence on substrate concentrations. This work investigates the possibility to approximate this dependency with artificial neural network. Also the influence of external diffusion layer to results of multi-substrate determination was investigated. The numerically modelled biosensor response was used as experimental data. The principal components analysis was used to reduce the dimension of biosensor response. Prefered method gives acceptable acuratnes on multi-substrate determination and it can be improved by relatively large external diffusion layer.Šiame darbe tiriama galimybė nustatyti kelių tirpale esančių medžiagų koncentracijas, naudojant biojutiklį kartu su dirbtiniu neuroniniu tinklu. Taip pat nagrinėjama kaip išorinis biojutiklio difuzijos sluoksnis įtakoja rezultatus. Darbe medžiagų koncentracijų prognozavimui naudojami ampermetrinio biojutiklio su išoriniu difuzijos sluoksniu kompiuterinio modeliavimo rezultatai. Biojutiklio atsako dimensiškumui mažinti naudota principinių komponenčių analizė. Principinės komponentės naudotos kaip dirbtinio neuroninio tinklo įeitis, nustatant tirpalo medžiagų koncentracijas. Gauti rezultatai pagrindžia dirbtinio neuroninio tinklo naudojimą šiam uždaviniui spręsti bei nusako išorinio difuzijos sluoksnio poveikį
- …
