TCC - Bacharelado em Ciência da Computação (Sede)
URI permanente para esta coleçãohttps://arandu.ufrpe.br/handle/123456789/415
Navegar
7 resultados
Resultados da Pesquisa
Item Development of machine learning models for the prediction of dissolved oxygen in aquaculture 4.0(2021-02-24) Freitas, Fábio Alves de; Nóbrega, Obionor de Oliveira; Lins, Fernando Antonio Aires; http://lattes.cnpq.br/2475965771605110; http://lattes.cnpq.br/8576087238071129; http://lattes.cnpq.br/5725435192607619The world faces the problem of feeding a growing population, which will reach more than 9 billion people by 2050. Thus, there is a need to develop activities that promote food production, within the dimensions of sustainability (social, technicaleconomic, and environmental). In this context, IoT systems focused on aquaculture 4.0 stand out, which allows the cultivation of high productions per unit of volume, with low environmental impact. However, these systems need to be extremely controlled, requiring sensors to perform realtime readings of water metrics, with emphasis on the dissolved oxygen (DO) sensor, which plays an essential role in determining the quality and quantity of available habitat for the organisms present in the system. Even with this importance, this sensor is often not used, due to its high associated cost. As an alternative solution to this problem, machine learning models have been proposed to predict DO, using temperature and pH readings as inputs. Experiments were carried out comparing different data scaling techniques and the prediction performance in different seasons of the year and regression metrics were used to evaluate the implemented models. The results showed that the proposed LSTM model is capable of making OD predictions and being applied in IoT and aquaculture 4.0 systems.Item Aprendizagem de máquina quântica e comitê quântico de classificadores(2019-12-02) Araujo, Ismael Cesar da Silva; Nascimento, André Câmara Alves do; Silva, Adenilton José da; http://lattes.cnpq.br/0314035098884256; http://lattes.cnpq.br/0622594061462533; http://lattes.cnpq.br/7125338940009959Quantum machine learning is a subarea of quantum computing that studies, among other things, the creation of equivalent classical classifiers. An ensemble of classifiers is a classification model in which the output is a combined result of the outputs of the classifiers contained in it. With the premiss that when using a sufficiently large ensemble with average classifiers, a good performance can still be obtained. This work investigates the differences in the performance of a quantum equivalent of an ensemble of classifiers, using trained and untrained classifiers. Where the simulation was mane, which the performance was measured through the calculation of the amplitude probabilities of the system. And the machine learning models of the ensemble were executed over benchmark datasets made available by scikitlearn library.Item Aprendizado profundo com capacidade computacional reduzida: uma aplicação à quebra de CAPTCHAs(2018-08-16) Melo, Diogo Felipe Félix de; Sampaio, Pablo Azevedo; http://lattes.cnpq.br/8865836949700771; http://lattes.cnpq.br/2213650736070295During the last decade, Deep Neural Networks has been shown to be a powerfull machine learn technique. Generally, to obtain relevant results, these techniques require high computacional power and large volumes of data, which can be a limiting factor on some cases. Neverthless, a careful project of trainig and archtecture may help to reduce these requirements. In the this work we present a comparative approach to the application of deep neural networks to text based CAPTCHAs as a way to cope with these limitations. We studied models that are capable of learn to segment and identify the text content of images, only based on examples. By experimentation of different hiper-parameters and architectures, we were capable to obtain a final model with 96.06% of token prediction accuracy in approximately 3 hours of training in a simple personal computer.Item Teacher-Student aplicado a diferentes modelos de recompensa do ambiente Lunar Lander(2021-07-20) Silva, Kenedy Felipe dos Santos da; Sampaio, Pablo Azevedo; http://lattes.cnpq.br/8865836949700771; http://lattes.cnpq.br/7479192156880225The techniques studied related to learning by reinforcement are becoming more and more common in real world challenges, but one challenge is to reduce the learning time. Currently, the learning time and/or amount of interactions performed by the reinforcement learning agent can result in high costs in applications, as the training of models can consume a lot of time, requiring many interactions between the agent and the task environment. This work seeks to improve learning using a new combination of techniques, the Teacher-Student technique with Reward Shaping. The Teacher-Student technique aims to choose among a set of similar tasks that train for a main task, according to the student’s learning. The Reward Shaping technique, altering the reward to try to accelerate learning, provides more frequent feedback on appropriate behaviors, that is, reports rewards more often. We adapted Teacher-Student algorithms for this combination of techniques, and used the Lunar Lander environment as a case study, using four reward models designed in (ALBUQUERQUE, 2021) for this environment. Experiments were performed running different trainings to compare this approach with training only on the original Lunar Lander (no rewards change), and with the results obtained in (ALBUQUERQUE, 2021) by adopting each of the rewards models individually. The combination of Teacher-Student techniques with Reward Shaping contributed to a new experience in the reinforcement learning area, managing to accelerate the agent’s learning, considering the duration of 600 thousand training steps, reaching the target performance in 2 out of 5 proposals, in addition to of being able to learn better than the original Lunar Lander approach with PPO algorithm.Item Prevendo a evasão escolar em uma instituição de ensino técnico utilizando mineração de dados educacionais(2021-07-16) Lemos, Ítalo Vinícius do Rego; Nascimento, André Câmara Alves do; http://lattes.cnpq.br/0622594061462533; http://lattes.cnpq.br/9146227756985212Dropping out of school is one of the main problems that occur in education and has become a very frequent reality within public or private educational institutions, resulting in immeasurable consequences both in the life of the student who fails to attend school and for society as a whole. Because it is a phenomenon that worries education professionals a lot, if it is necessary to reverse it, so these professionals need resources that are efficient in demonstrating knowledge inside and outside the teaching environment and outlining strategies to deal with such a scenario. Being able to predict a possible dropout benefits both the student and institutions. Based on this, a methodology that has proven to be skillful in combating school dropouts and capable of providing knowledge to the educational institution is Educational Data Mining. Based on this, this work aimed to apply Data Mining and Machine Learning techniques to predict possible dropout cases before the student enters the educational institution. Through social and economic indicators of the student and his family, he is classified as a potential evader or not. This study adopted a real database from a Brazilian public education institution, with data from candidates who competed in its admission process (entrance exam) for a place in technical education. During the research, 3 classification models were used Decision Tree, Random Forest e XGBoost with the XGBoost algorithm achieving a 74% hit rate in predicting evaders, being superior to the others but still presenting a high number of students classified as not dropouts but who actually dropped out. Given these results, we conclude that more indicators are needed to satisfactorily detect the possible candidate who will drop out.Item Inferência automática de nível de dificuldade de receitas culinárias usando técnicas de processamento de linguagem natural(2020-12-21) Britto, Larissa Feliciana da Silva; Pacífico, Luciano Demétrio Santos; Ludermir, Teresa Bernarda; http://lattes.cnpq.br/6321179168854922; http://lattes.cnpq.br/9521600706234665; http://lattes.cnpq.br/5058497100007411In this work, a tool for inferring the degree of difficulty of cooking recipes will be proposed. The inference will be made by the textual classification of the recipe preparation methods. The tool will be a fundamental piece to the development of a contextaware contentbased cooking recipe recommendation system. Some of the main classifiers in Text Classification literature will be adopted, in addition to different feature extraction methods. An experimental evaluation is performed, in order to select the best approaches to compose the system.Item Tratamento de Kernels incompletos em redes bipartidas na predição de interações em redes biológicas(2020-10-30) Bastos, Tássia Laís Barros; Nascimento, André Câmara Alves do; http://lattes.cnpq.br/0622594061462533; http://lattes.cnpq.br/2065961687962702In the last decade, the study of pharmacological networks has received a lot of attention given its relevance to the production of new drugs. The studies were made possible by the large volume of biological data generated, making it possible to understand and extract knowledge from them. However, although interesting, this is a process that brings with it some barriers in terms of viability, particularly when the data appear heterogeneously and contain missing information. Many different approaches for predicting biological interactions have been proposed, especially in the area of multiple kernel learning (Multiple Kernel Learning (MKL). The use of MKL methods in biological data is also compromised by the heterogeneity of data sources, but associated with the methods, techniques for complementing missing values in the base kernel matrices can be used, this filling process is usually done with simple techniques, such as imputing zeroes, mean and median of the matrix. In this work, techniques for handling false values were evaluated in the context of bipartite networks to solve the limitations related to the heterogeneity of the data. We used three single value imputation techniques (mean, median and zero) and a more complex predictive imputation technique (SVD). All the aforementioned techniques have already been used for matrix completeness in the context of unipartite networks. Our analyzes showed that the SVD technique performed much better than the other techniques in evaluative metrics, bringing encouraging results for the use of the technique in models that use bipartite networks. The average and median techniques showed similar performances, but lower than the SVD. And filling with zero showed the worst performance in relation to all other applied techniques.