在本論文中，我們提出了一個以規則為基礎的分類演算法，名為ROUSER（ROUgh SEt Rule），它利用粗糙集理論作為搜尋啟發的基礎，進而建立規則。我們使用一個已經被廣泛利用的工具實作ROUSER，也使用數個公開資料集對它進行實驗，並將它應用於真實世界的案例。 本論文的初衷可被追溯到一個真實世界的案例，而此案例的目標是從感應器所蒐集的資料中找出與機械故障之間的關聯。為了能支援機械故障的根本原因分析，我們設計並實作了一個以規則為基礎的分類演算法，它所產生的模型是由人類可理解的決策規則所組成，而故障的徵兆與原因則被決策規則所連結。此外，資料中存在著矛盾。舉例而言，不同時間點所蒐集的兩筆紀錄極為相似、甚至相同（除了時間戳記），但其中一筆紀錄與機械故障相關，另一筆則否。本案例的挑戰在於分析矛盾的資料。我們使用粗糙集理論克服這個難題，因為它可以處理不完美知識。 研究者們已經提出了各種不同的分類演算法，而實踐者們則已經將它們應用於各種領域，然而多數分類演算法的設計並不強調演算法所產生模型的可解釋性與可理解性。ROUSER的設計是專門從名目資料中萃取人類可理解的決策規則。而ROUSER與其它多數規則分類演算法不同的地方是利用粗糙集方法選取特徵。ROUSER也提供了數種方式來選擇合宜的屬性與值配對，作為規則的前項。此外，ROUSER的規則產生方法是基於separate-and-conquer策略，因此比其它基於粗糙集的分類演算法所廣泛採用的不可分辨矩陣方法還有效率。 我們進行延伸實驗來驗證ROUSER的能力。對於名目資料的實驗裡，ROUSER在半數的結果中的準確率可匹敵、甚至勝過其他以規則為基礎的分類演算法以及決策樹分類演算法。ROUSER也可以在一些離散化的資料集之中達到可匹敵甚至超越的準確率。我們也提供了內建的特徵萃取方法與其它方法的比較的實驗結果，以及數種用來決定規則前項的方法的實驗結果。 / In this thesis, we propose a rule-based classification algorithm named ROUSER (ROUgh SEt Rule), which uses the rough set theory as the basis of the search heuristics in the process of rule generation. We implement ROUSER using a well developed and widely used toolkit, evaluate it using several public data sets, and examine its applicability using a real-world case study. The origin of the problem addressed in this thesis can be traced back to a real-world problem where the goal is to determine whether a data record collected from a sensor corresponds to a machine fault. In order to assist in the root cause analysis of the machine faults, we design and implement a rule-based classification algorithm that can generate models consisting of human understandable decision rules to connect symptoms to the cause. Moreover, there are contradictions in data. For example, two data records collected at different time points are similar, or the same (except their timestamps), while one is corresponding to a machine fault but not the other. The challenge is to analyze data with contradictions. We use the rough set theory to overcome the challenge, since it is able to process imperfect knowledge. Researchers have proposed various classification algorithms and practitioners have applied them to various application domains, while most of the classification algorithms are designed without a focus on interpretability or understandability of the models built using the algorithms. ROUSER is specifically designed to extract human understandable decision rules from nominal data. What distinguishes ROUSER from most, if not all, other rule-based classification algorithms is that it utilizes a rough set approach to select features. ROUSER also provides several ways to decide an appropriate attribute-value pair for the antecedents of a rule. Moreover, the rule generation method of ROUSER is based on the separate-and-conquer strategy, and hence it is more efficient than the indiscernibility matrix method that is widely adopted in the classification algorithms based on the rough set theory. We conduct extensive experiments to evaluate the capability of ROUSER. On about half of the nominal data sets considered in experiments, ROUSER can achieve comparable or better accuracy than do classification algorithms that are able to generate decision rules or trees. On some of the discretized data sets, ROUSER can achieve comparable or better accuracy. We also present the results of the experiments on the embedded feature selection method and several ways to decide an appropriate attribute-value pair for the antecedents of a rule.

The present dissertation set out to investigate how the linguistic environment affects speech perception. Three sets of studies have explored effects of bilingualism on word recognition in adults and infants and the impact of first language linguistic knowledge on rule learning in adults. In the present work, we have found evidence in three auditory priming studies that bilingual adults, in contrast to monolinguals have developed mechanisms to effectively overcome interference from irrelevant information in the speech signal. Preliminary results on toddlers indicate no differences in the recognition of mispronounced words between bilinguals and monolinguals. Additionally, knowledge about rules in the first language was shown to have an impact on general rule learning abilities, while we did not detect an influence of bilingualism in this process. In summary, we have found evidence for an impact of the linguistic environment on the processing of indexical variability in word recognition as well as on rule learning. Bilinguals seem to have adapted to increased variability in their daily speech environment. In addition, rule extraction from unknown language input was unaffected by those adaptations but influenced by linguistic knowledge. / El principal objetivo de este trabajo fue investigar cómo el entorno lingüístico afecta la percepción del habla. Con esta finalidad se exploraron los efectos del bilingüismo en el reconocimiento de palabras y el efecto de la lengua materna en el aprendizaje de reglas. Para ver los efectos del bilingüismo en el reconocimiento de palabras se realizaron tres estudios en adultos explorando el efecto de facilitación por repetición, que mostraron que las personas bilingües han desarrollado mecanismos que les permiten minimizar las interferencias que ejerce la información irrelevante en la señal del habla. Por otro lado, se realizó un estudio con niños pequeños cutos resultados sugieren que no hay diferencias en el reconocimiento de palabras mal pronunciadas entre niños pequeños bilingües y monolingües. Respecto al efecto del conocimiento lingüístico de la lengua materna en el aprendizaje de reglas, se mostró que tiene un impacto en las habilidades generales para el aprendizaje de reglas aunque no se ha detectó una influencia del bilingüismo en dicho proceso. En resumen, se ha mostrado que el bilingüismo minimiza los efectos negativos de la variabilidad en el reconocimiento de palabras. Los bilingües parecen haberse adaptado a una mayor variabilidad en su entorno de habla cotidiana. Por el otro lado, se ha visto que la capacidad para extraer reglas de una lengua desconocida no está afectada por estas adaptaciones si no que está influenciada por los conocimientos lingüísticos en la lengua materna.

Os cartões, sejam de crédito ou débito, são meios de pagamento altamente utilizados. Esse fato desperta o interesse de fraudadores. O mercado de cartões enxerga as fraudes como custos operacionais, que são repassados para os consumidores e para a sociedade em geral. Ainda, o alto volume de transações e a necessidade de combater as fraudes abrem espaço para a aplicação de técnicas de Aprendizagem de Máquina; entre elas, os classificadores. Um tipo de classificador largamente utilizado nesse domínio é o classificador baseado em regras. Entretanto, um ponto de atenção dessa categoria de classificadores é que, na prática, eles são altamente dependentes dos especialistas no domínio, ou seja, profissionais que detectam os padrões das transações fraudulentas, os transformam em regras e implementam essas regras nos sistemas de classificação. Ao reconhecer esse cenário, o objetivo desse trabalho é propor a uma arquitetura baseada em regras de associação e regressão logística - técnicas estudadas em Aprendizagem de Máquina - para minerar regras nos dados e produzir, como resultado, conjuntos de regras de detecção de transações fraudulentas e disponibilizá-los para os especialistas no domínio. Com isso, esses profissionais terão o auxílio dos computadores para descobrir e gerar as regras que embasam o classificador, diminuindo, então, a chance de haver padrões fraudulentos ainda não reconhecidos e tornando as atividades de gerar e manter as regras mais eficientes. Com a finalidade de testar a proposta, a parte experimental do trabalho contou com cerca de 7,7 milhões de transações reais de cartões fornecidas por uma empresa participante do mercado de cartões. A partir daí, dado que o classificador pode cometer erros (falso-positivo e falso-negativo), a técnica de análise sensível ao custo foi aplicada para que a maior parte desses erros tenha um menor custo. Além disso, após um longo trabalho de análise do banco de dados, 141 características foram combinadas para, com o uso do algoritmo FP-Growth, gerar 38.003 regras que, após um processo de filtragem e seleção, foram agrupadas em cinco conjuntos de regras, sendo que o maior deles tem 1.285 regras. Cada um desses cinco conjuntos foi submetido a uma modelagem de regressão logística para que suas regras fossem validadas e ponderadas por critérios estatísticos. Ao final do processo, as métricas de ajuste estatístico dos modelos revelaram conjuntos bem ajustados e os indicadores de desempenho dos classificadores também indicaram, num geral, poderes de classificação muito bons (AROC entre 0,788 e 0,820). Como conclusão, a aplicação combinada das técnicas estatísticas - análise sensível ao custo, regras de associação e regressão logística - se mostrou conceitual e teoricamente coesa e coerente. Por fim, o experimento e seus resultados demonstraram a viabilidade técnica e prática da proposta. / Credit and debit cards are two methods of payments highly utilized. This awakens the interest of fraudsters. Businesses see fraudulent transactions as operating costs, which are passed on to consumers. Thus, the high number of transactions and the necessity to combat fraud stimulate the use of machine learning algorithms; among them, rule-based classifiers. However, a weakness of these classifiers is that, in practice, they are highly dependent on professionals who detect patterns of fraudulent transactions, transform them into rules and implement these rules in the classifier. Knowing this scenario, the aim of this thesis is to propose an architecture based on association rules and logistic regression - techniques studied in Machine Learning - for mining rules on data and produce rule sets to detect fraudulent transactions and make them available to experts. As a result, these professionals will have the aid of computers to discover the rules that support the classifier, decreasing the chance of having non-discovered fraudulent patterns and increasing the efficiency of generate and maintain these rules. In order to test the proposal, the experimental part of the thesis has used almost 7.7 million transactions provided by a real company. Moreover, after a long process of analysis of the database, 141 characteristics were combined using the algorithm FP-Growth, generating 38,003 rules. After a process of filtering and selection, they were grouped into five sets of rules which the biggest one has 1,285 rules. Each of the five sets was subjected to logistic regression, so their rules have been validated and weighted by statistical criteria. At the end of the process, the goodness of fit tests were satisfied and the performance indicators have shown very good classification powers (AUC between 0.788 and 0.820). In conclusion, the combined application of statistical techniques - cost sensitive learning, association rules and logistic regression - proved being conceptually and theoretically cohesive and coherent. Finally, the experiment and its results have demonstrated the technical and practical feasibilities of the proposal.

Warehouse automation is a measure E-commerce companies can take to get a more streamlined flow through their warehouse. Order picking is the most labor intensive task in a warehouse. By automating the order picking process companies can lower their costs and improve their response times. This thesis studies the A-frame, an automated order picking system, at a large online pharmacy, Apotea AB. An A-frame has dispensing channels on its side and a conveyor belt that runs through the entire machine. Products for an order are ejected from the channels onto the conveyor belt and at the end of the machine they are dropped into a box. The box is then sealed, labeled and sent to the customer. For the automatic flow to function correctly, all orders picked by the A-frame need to be complete orders. Complete orders are orders where there are no products missing. To maximize the throughput of the A-frame, an appropriate product allocation will be required. Due to the vast number of combinations, it is extremely difficult to identify an optimal product allocation. This study has examined three different approaches to the product allocation problem for an A-frame. The first two methods are based on ranking the products depending on their quantities sold. The last method uses association rule learning, which is a machine learning technique for finding interesting patterns in a data set. Association rule learning was used to find which products were associated to each other. These associations were then placed in a graph structure and solved using a heuristic. To evaluate the different allocation methods, a simulation model was created. The A-frame was simulated using a discrete event simulation, which meant all methods could be tested on the same data to correctly compare the performance of each allocation. The study showed that the heuristic using association rules gave the highest number of picks for the tested period. However, it was only marginally better than the method that first removed orders that could not be picked from the A-frame and then ranked all products by their quantities sold. The study's conclusion is that while association rule learning resulted in the highest number of picked orders, the gain of using it does not motivate its complexity. Instead a more simple approach by ranking products by their quantities sold should be used. Warehousing in the era of E-commerce has to be fast, correct and cheap.

Off-the-shelf Reinforcement Learning (RL) algorithms suffer from slow learning performance, partly because they are expected to learn a task from scratch merely through an agent's own experience. In this thesis, we show that learning from scratch is a limiting factor for the learning performance, and that when prior knowledge is available RL agents can learn a task faster. We evaluate relevant previous work and our own algorithms in various experiments. Our first contribution is the first implementation and evaluation of an existing interactive RL algorithm in a real-world domain with a humanoid robot. Interactive RL was evaluated in a simulated domain which motivated us for evaluating its practicality on a robot. Our evaluation shows that guidance reduces learning time, and that its positive effects increase with state space size. A natural follow up question after our first evaluation was, how do some other previous works compare to interactive RL. Our second contribution is an analysis of a user study, where na"ive human teachers demonstrated a real-world object catching with a humanoid robot. We present the first comparison of several previous works in a common real-world domain with a user study. One conclusion of the user study was the high potential of RL despite poor usability due to slow learning rate. As an effort to improve the learning efficiency of RL learners, our third contribution is a novel human-agent knowledge transfer algorithm. Using demonstrations from three teachers with varying expertise in a simulated domain, we show that regardless of the skill level, human demonstrations can improve the asymptotic performance of an RL agent. As an alternative approach for encoding human knowledge in RL, we investigated the use of reward shaping. Our final contributions are Static Inverse Reinforcement Learning Shaping and Dynamic Inverse Reinforcement Learning Shaping algorithms that use human demonstrations for recovering a shaping reward function. Our experiments in simulated domains show that our approach outperforms the state-of-the-art in cumulative reward, learning rate and asymptotic performance. Overall we show that human demonstrators with varying skills can help RL agents to learn tasks more efficiently.

Made available in DSpace on 2016-06-02T19:05:54Z (GMT). No. of bitstreams: 1 3998.pdf: 3486824 bytes, checksum: f1c040adfdc7d0672bc93a058f8a413d (MD5) Previous issue date: 2011-06-28 / Financiadora de Estudos e Projetos / The goal of this work is to study, expand and evaluate the use of multiobjective genetic algorithms and the iterative rule learning approach in fuzzy system generation, especially, in fuzzy rule-based systems, both in automatic fuzzy rule generation from datasets and in fuzzy sets optimization. This work investigates the use of multi-objective genetic algorithms with a focus on the trade-off between accuracy and interpretability, considered contradictory objectives in the representation of fuzzy systems. With this purpose, we propose and implement an evolutive multi-objective genetic model composed of three stages. In the first stage uniformly distributed fuzzy sets are created. In the second stage, the rule base is generated by using an iterative rule learning approach and a multiobjective genetic algorithm. Finally the fuzzy sets created in the first stage are optimized through a multi-objective genetic algorithm. The proposed model was evaluated with a number of benchmark datasets and the results were compared to three other methods found in the literature. The results obtained with the optimization of the fuzzy sets were compared to the result of another fuzzy set optimizer found in the literature. Statistical comparison methods usually applied in similar context show that the proposed method has an improved classification rate and interpretability in comparison with the other methods. / O objetivo deste trabalho é estudar, expandir e avaliar o uso dos algoritmos genéticos multiobjetivo e a abordagem iterativa na geração de sistemas fuzzy, mais especificamente para sistemas fuzzy baseados em regras, tanto na geração automática da base de regras fuzzy a partir de conjuntos de dados, como a otimização dos conjuntos fuzzy. Esse trabalho investiga o uso dos algoritmos genéticos multiobjetivo com enfoque na questão de balanceamento entre precisão e interpretabilidade, ambos considerados contraditórios entre si na representação de sistemas fuzzy. Com este intuito, é proposto e implementado um modelo evolutivo multiobjetivo genético composto por três etapas. Na primeira etapa são criados os conjuntos fuzzy uniformemente distribuídos. Na segunda etapa é tratada a geração da base de regras usando a abordagem iterativa e um algoritmo genético multiobjetivo. Por fim, na terceira etapa os conjuntos fuzzy criados na primeira etapa são otimizados mediante um algoritmo genético multiobjetivo. O modelo desenvolvido foi avaliado em diversos conjuntos de dados benchmark e os resultados obtidos foram comparados com outros três métodos, que geram regras de classificação, encontrados na literatura. Os resultados obtidos após a otimização dos conjuntos fuzzy foram comparados com resultados de outro otimizador de conjuntos fuzzy encontrado na literatura. Métodos estatísticos de comparação usualmente aplicados em contextos semelhantes mostram uma melhor taxa de classificação e interpretabilidade do método proposto com relação a outros métodos.

No description available.

At the beginning, the work describes DNS and SSL/TLS protocols, it mainly deals with communication between devices using these protocols. Then we'll talk about data preprocessing and data cleaning. Furthermore, the thesis deals with basic data mining techniques such as data classification, association rules, information retrieval, regression analysis and cluster analysis. The next chapter we can read something about how to identify mobile devices on the network. We will evaluate data sets that contain collected data from communication between the above mentioned protocols, which will be used in the practical part. After that, we finally get to the design of a system for analyzing network communication data. We will describe the libraries, which we used and the entire system implementation. We will perform a large number of experiments, which we will finally evaluate.