Recentemente, houve um aumento rápido na criação e disponibilidade de repositórios de dados, o que foi percebido nas áreas de Mineração de Dados e Aprendizagem de Máquina. Este fato deve-se principalmente à rápida criação de tais dados em redes sociais. Uma grande parte destes dados é feita de texto, e a informação armazenada neles pode descrever desde perfis de usuários a temas comuns em documentos como política, esportes e ciência, informação bastante útil para várias aplicações. Como muitos destes dados são criados em fluxos, é desejável a criação de algoritmos com capacidade de atuar em grande escala e também de forma on-line, já que tarefas como organização e exploração de grandes coleções de dados seriam beneficiadas por eles. Nesta dissertação um modelo probabilístico, on-line e incremental é apresentado, como um esforço em resolver o problema apresentado. O algoritmo possui o nome DV-INBC e é uma extensão ao algoritmo INBC. As duas principais características do DV-INBC são: a necessidade de apenas uma iteração pelos dados de treino para criar um modelo que os represente; não é necessário saber o vocabulário dos dados a priori. Logo, pouco conhecimento sobre o fluxo de dados é necessário. Para avaliar a performance do algoritmo, são apresentados testes usando datasets populares. / Recently the fields of Data Mining and Machine Learning have seen a rapid increase in the creation and availability of data repositories. This is mainly due to its rapid creation in social networks. Also, a large part of those data is made of text documents. The information stored in such texts can range from a description of a user profile to common textual topics such as politics, sports and science, information very useful for many applications. Besides, since many of this data are created in streams, scalable and on-line algorithms are desired, because tasks like organization and exploration of large document collections would be benefited by them. In this thesis an incremental, on-line and probabilistic model for document classification is presented, as an effort of tackling this problem. The algorithm is called DV-INBC and is an extension to the INBC algorithm. The two main characteristics of DV-INBC are: only a single scan over the data is necessary to create a model of it; the data vocabulary need not to be known a priori. Therefore, little knowledge about the data stream is needed. To assess its performance, tests using well known datasets are presented.

Classificação de fluxos contínuos de dados possui muitos desafios para a comunidade de mineração de dados quando o ambiente não é estacionário. Um dos maiores desafios para a aprendizagem em fluxos contínuos de dados está relacionado com a adaptação às mudanças de conceito, as quais ocorrem como resultado da evolução dos dados ao longo do tempo. Duas formas principais de desenvolver abordagens adaptativas são os métodos baseados em conjunto de classificadores e os algoritmos incrementais. Métodos baseados em conjunto de classificadores desempenham um papel importante devido à sua modularidade, o que proporciona uma maneira natural de se adaptar a mudanças de conceito. Os algoritmos incrementais são mais rápidos e possuem uma melhor capacidade anti-ruído do que os conjuntos de classificadores, mas têm mais restrições sobre os fluxos de dados. Assim, é um desafio combinar a flexibilidade e a adaptação de um conjunto de classificadores na presença de mudança de conceito, com a simplicidade de uso encontrada em um único classificador com aprendizado incremental. Com essa motivação, nesta dissertação, propomos um algoritmo incremental, online e probabilístico para a classificação em problemas que envolvem mudança de conceito. O algoritmo é chamado IGMN-NSE e é uma adaptação do algoritmo IGMN. As duas principais contribuições da IGMN-NSE em relação à IGMN são: melhoria de poder preditivo para tarefas de classificação e a adaptação para alcançar um bom desempenho em cenários não estacionários. Estudos extensivos em bases de dados sintéticas e do mundo real demonstram que o algoritmo proposto pode rastrear os ambientes em mudança de forma muito próxima, independentemente do tipo de mudança de conceito. / Data stream classification poses many challenges for the data mining community when the environment is non-stationary. The greatest challenge in learning classifiers from data stream relates to adaptation to the concept drifts, which occur as a result of changes in the underlying concepts. Two main ways to develop adaptive approaches are ensemble methods and incremental algorithms. Ensemble method plays an important role due to its modularity, which provides a natural way of adapting to change. Incremental algorithms are faster and have better anti-noise capacity than ensemble algorithms, but have more restrictions on concept drifting data streams. Thus, it is a challenge to combine the flexibility and adaptation of an ensemble classifier in the presence of concept drift, with the simplicity of use found in a single classifier with incremental learning. With this motivation, in this dissertation we propose an incremental, online and probabilistic algorithm for classification as an effort of tackling concept drifting. The algorithm is called IGMN-NSE and is an adaptation of the IGMN algorithm. The two main contributions of IGMN-NSE in relation to the IGMN are: predictive power improvement for classification tasks and adaptation to achieve a good performance in non-stationary environments. Extensive studies on both synthetic and real-world data demonstrate that the proposed algorithm can track the changing environments very closely, regardless of the type of concept drift.

Classificação de fluxos contínuos de dados possui muitos desafios para a comunidade de mineração de dados quando o ambiente não é estacionário. Um dos maiores desafios para a aprendizagem em fluxos contínuos de dados está relacionado com a adaptação às mudanças de conceito, as quais ocorrem como resultado da evolução dos dados ao longo do tempo. Duas formas principais de desenvolver abordagens adaptativas são os métodos baseados em conjunto de classificadores e os algoritmos incrementais. Métodos baseados em conjunto de classificadores desempenham um papel importante devido à sua modularidade, o que proporciona uma maneira natural de se adaptar a mudanças de conceito. Os algoritmos incrementais são mais rápidos e possuem uma melhor capacidade anti-ruído do que os conjuntos de classificadores, mas têm mais restrições sobre os fluxos de dados. Assim, é um desafio combinar a flexibilidade e a adaptação de um conjunto de classificadores na presença de mudança de conceito, com a simplicidade de uso encontrada em um único classificador com aprendizado incremental. Com essa motivação, nesta dissertação, propomos um algoritmo incremental, online e probabilístico para a classificação em problemas que envolvem mudança de conceito. O algoritmo é chamado IGMN-NSE e é uma adaptação do algoritmo IGMN. As duas principais contribuições da IGMN-NSE em relação à IGMN são: melhoria de poder preditivo para tarefas de classificação e a adaptação para alcançar um bom desempenho em cenários não estacionários. Estudos extensivos em bases de dados sintéticas e do mundo real demonstram que o algoritmo proposto pode rastrear os ambientes em mudança de forma muito próxima, independentemente do tipo de mudança de conceito. / Data stream classification poses many challenges for the data mining community when the environment is non-stationary. The greatest challenge in learning classifiers from data stream relates to adaptation to the concept drifts, which occur as a result of changes in the underlying concepts. Two main ways to develop adaptive approaches are ensemble methods and incremental algorithms. Ensemble method plays an important role due to its modularity, which provides a natural way of adapting to change. Incremental algorithms are faster and have better anti-noise capacity than ensemble algorithms, but have more restrictions on concept drifting data streams. Thus, it is a challenge to combine the flexibility and adaptation of an ensemble classifier in the presence of concept drift, with the simplicity of use found in a single classifier with incremental learning. With this motivation, in this dissertation we propose an incremental, online and probabilistic algorithm for classification as an effort of tackling concept drifting. The algorithm is called IGMN-NSE and is an adaptation of the IGMN algorithm. The two main contributions of IGMN-NSE in relation to the IGMN are: predictive power improvement for classification tasks and adaptation to achieve a good performance in non-stationary environments. Extensive studies on both synthetic and real-world data demonstrate that the proposed algorithm can track the changing environments very closely, regardless of the type of concept drift.

Recentemente, houve um aumento rápido na criação e disponibilidade de repositórios de dados, o que foi percebido nas áreas de Mineração de Dados e Aprendizagem de Máquina. Este fato deve-se principalmente à rápida criação de tais dados em redes sociais. Uma grande parte destes dados é feita de texto, e a informação armazenada neles pode descrever desde perfis de usuários a temas comuns em documentos como política, esportes e ciência, informação bastante útil para várias aplicações. Como muitos destes dados são criados em fluxos, é desejável a criação de algoritmos com capacidade de atuar em grande escala e também de forma on-line, já que tarefas como organização e exploração de grandes coleções de dados seriam beneficiadas por eles. Nesta dissertação um modelo probabilístico, on-line e incremental é apresentado, como um esforço em resolver o problema apresentado. O algoritmo possui o nome DV-INBC e é uma extensão ao algoritmo INBC. As duas principais características do DV-INBC são: a necessidade de apenas uma iteração pelos dados de treino para criar um modelo que os represente; não é necessário saber o vocabulário dos dados a priori. Logo, pouco conhecimento sobre o fluxo de dados é necessário. Para avaliar a performance do algoritmo, são apresentados testes usando datasets populares. / Recently the fields of Data Mining and Machine Learning have seen a rapid increase in the creation and availability of data repositories. This is mainly due to its rapid creation in social networks. Also, a large part of those data is made of text documents. The information stored in such texts can range from a description of a user profile to common textual topics such as politics, sports and science, information very useful for many applications. Besides, since many of this data are created in streams, scalable and on-line algorithms are desired, because tasks like organization and exploration of large document collections would be benefited by them. In this thesis an incremental, on-line and probabilistic model for document classification is presented, as an effort of tackling this problem. The algorithm is called DV-INBC and is an extension to the INBC algorithm. The two main characteristics of DV-INBC are: only a single scan over the data is necessary to create a model of it; the data vocabulary need not to be known a priori. Therefore, little knowledge about the data stream is needed. To assess its performance, tests using well known datasets are presented.

L'apprentissage statistique propose un vaste ensemble de techniques capables de construire des modèles prédictifs à partir d'observations passées. Ces techniques ont montré leurs capacités à traiter des volumétries importantes de données sur des problèmes réels. Cependant, de nouvelles applications génèrent de plus en plus de données qui sont seulement visibles sous la forme d'un flux et doivent être traitées séquentiellement. Parmi ces applications on citera : la gestion de réseaux de télécommunications, la modélisation des utilisateurs au sein d'un réseau social, le web mining. L'un des défis techniques est de concevoir des algorithmes permettant l'apprentissage avec les nouvelles contraintes imposées par les flux de données. Nous proposons d'abord ce problème en proposant de nouvelles techniques de résumé de flux de données dans le cadre de l'apprentissage supervisé. Notre méthode est constituée de deux niveaux. Le premier niveau utilise des techniques incrémentales de résumé en-ligne pour les flux qui prennent en compte les ressources mémoire et processeur et possèdent des garanties en termes d'erreur. Le second niveau utilise les résumés de faible taille, issus du premier niveau, pour construire le résumé final à l'aide d'une méthode supervisée performante hors-ligne. Ces résumés constituent un prétraitement qui nous permet de proposer de nouvelles versions du classifieur bayésien naïf et des arbres de décision fonctionnant en-ligne sur flux de données. Les flux de données peuvent ne pas être stationnaires mais comporter des changements de concept. Nous proposons aussi une nouvelle technique pour détecter ces changements et mettre à jour nos classifieurs. / Statistical learning provides numerous algorithms to build predictive models on past observations. These techniques proved their ability to deal with large scale realistic problems. However, new domains generate more and more data which are only visible once and need to be processes sequentially. These volatile data, known as data streams, come from telecommunication network management, social network, web mining. The challenge is to build new algorithms able to learn under these constraints. We proposed to build new summaries for supervised classification. Our summaries are based on two levels. The first level is an online incremental summary which uses low processing and address the precision/memory tradeoff. The second level uses the first layer summary to build the final sumamry with an effcient offline method. Building these sumamries is a pre-processing stage to develop new classifiers for data streams. We propose new versions for the naive-Bayes and decision trees classifiers using our summaries. As data streams might contain concept drifts, we also propose a new technique to detect these drifts and update classifiers accordingly.

abstract: In the last decade deep learning based models have revolutionized machine learning and computer vision applications. However, these models are data-hungry and training them is a time-consuming process. In addition, when deep neural networks are updated to augment their prediction space with new data, they run into the problem of catastrophic forgetting, where the model forgets previously learned knowledge as it overfits to the newly available data. Incremental learning algorithms enable deep neural networks to prevent catastrophic forgetting by retaining knowledge of previously observed data while also learning from newly available data. This thesis presents three models for incremental learning; (i) Design of an algorithm for generative incremental learning using a pre-trained deep neural network classifier; (ii) Development of a hashing based clustering algorithm for efficient incremental learning; (iii) Design of a student-teacher coupled neural network to distill knowledge for incremental learning. The proposed algorithms were evaluated using popular vision datasets for classification tasks. The thesis concludes with a discussion about the feasibility of using these techniques to transfer information between networks and also for incremental learning applications. / Dissertation/Thesis / Masters Thesis Computer Science 2020

L'intelligence artificielle est la discipline de recherche d'imitation ou de remplacement de fonctions cognitives humaines. À ce titre, l'une de ses branches s'inscrit dans l'automatisation progressive du processus de programmation. Il s'agit alors de transférer de l'intelligence ou, à défaut de définition, de transférer de la charge cognitive depuis l'humain vers le système, qu'il soit autonome ou guidé par l'utilisateur. Dans le cadre de cette thèse, nous considérons les conditions de l'évolution depuis un système guidé par son utilisateur vers un système autonome, en nous appuyant sur une autre branche de l'intelligence artificielle : l'apprentissage artificiel. Notre cadre applicatif est celui de la conception d'un assistant opérationnel incrémental, c'est-à-dire d'un système capable de réagir à des requêtes formulées par l'utilisateur en adoptant les actions appropriées, et capable d'apprendre à le faire. Pour nos travaux, les requêtes sont exprimées en français, et les actions sont désignées par les commandes correspondantes dans un langage de programmation (ici, R ou bash). L'apprentissage du système est effectué à l'aide d'un ensemble d'exemples constitué par les utilisateurs eux-mêmes lors de leurs interactions. Ce sont donc ces derniers qui définissent, progressivement, les actions qui sont appropriées pour chaque requête, afin de rendre le système de plus en plus autonome. Nous avons collecté plusieurs ensembles d'exemples pour l'évaluation des méthodes d'apprentissage, en analysant et réduisant progressivement les biais induits. Le protocole que nous proposons est fondé sur l'amorçage incrémental des connaissances du système à partir d'un ensemble vide ou très restreint. Cela présente l'avantage de constituer une base de connaissances très représentative des besoins des utilisateurs, mais aussi l'inconvénient de n'aquérir qu'un nombre très limité d'exemples. Nous utilisons donc, après examen des performances d'une méthode naïve, une méthode de raisonnement à partir de cas : le raisonnement par analogie formelle. Nous montrons que cette méthode permet une précision très élevée dans les réponses du système, mais également une couverture relativement faible. L'extension de la base d'exemples par analogie est explorée afin d'augmenter la couverture des réponses données. Dans une autre perspective, nous explorons également la piste de rendre l'analogie plus tolérante au bruit et aux faibles différences en entrée en autorisant les approximations, ce qui a également pour effet la production de réponses incorrectes plus nombreuses. La durée d'exécution de l'approche par analogie, déjà de l'ordre de la seconde, souffre beaucoup de l'extension de la base et de l'approximation. Nous avons exploré plusieurs méthodes de segmentation des séquences en entrée afin de réduire cette durée, mais elle reste encore le principal obstacle à contourner pour l'utilisation de l'analogie formelle dans le traitement automatique de la langue. Enfin, l'assistant opérationnel incrémental fondé sur le raisonnement analogique a été testé en condition incrémentale simulée, afin d'étudier la progression de l'apprentissage du système au cours du temps. On en retient que le modèle permet d'atteindre un taux de réponse stable après une dizaine d'exemples vus en moyenne pour chaque type de commande. Bien que la performance effective varie selon le nombre total de commandes considérées, cette propriété ouvre sur des applications intéressantes dans le cadre incrémental du transfert depuis un domaine riche (la langue naturelle) vers un domaine moins riche (le langage de programmation). / Artificial Intelligence is the field of research aiming at mimicking or replacing human cognitive abilities. As such, one of its subfields is focused on the progressive automation of the programming process. In other words, the goal is to transfer cognitive load from the human to the system, whether it be autonomous or guided by the user. In this thesis, we investigate the conditions for making a user-guided system autonomous using another subfield of Artificial Intelligence : Machine Learning. As an implementation framework, we chose the design of an incremental operational assistant, that is a system able to react to natural language requests from the user with relevant actions. The system must also be able to learn the correct reactions, incrementally. In our work, the requests are in written French, and the associated actions are represented by corresponding instructions in a programming language (here R and bash). The learning is performed using a set of examples composed by the users themselves while interacting. Thus they progressively define the most relevant actions for each request, making the system more autonomous. We collected several example sets for evaluation of the learning methods, analyzing and reducing the inherent collection biases. The proposed protocol is based on incremental bootstrapping of the system, starting from an empty or limited knowledge base. As a result of this choice, the obtained knowledge base reflects the user needs, the downside being that the overall number of examples is limited. To avoid this problem, after assessing a baseline method, we apply a case base reasoning approach to the request to command transfer problem: formal analogical reasoning. We show that this method yields answers with a very high precision, but also a relatively low coverage. We explore the analogical extension of the example base in order to increase the coverage of the provided answers. We also assess the relaxation of analogical constraints for an increased tolerance of analogical reasoning to noise in the examples. The running delay of the simple analogical approach is already around 1 second, and is badly influenced by both the automatic extension of the base and the relaxation of the constraints. We explored several segmentation strategies on the input examples in order to reduce reduce this time. The delay however remains the main obstacle to using analogical reasoning for natural language processing with usual volumes of data. Finally, the incremental operational assistant based on analogical reasoning was tested in simulated incremental condition in order to assess the learning behavior over time. The system reaches a stable correct answer rate after a dozen examples given in average for each command type. Although the effective performance depends on the total number of accounted commands, this observation opens interesting applicative tracks for the considered task of transferring from a rich source domain (natural language) to a less rich target domain (programming language).

This thesis concerns the design, implementation and evaluation of a specification based testing architecture for reactive systems using the paradigm of learning-based testing. As part of this work we have designed, verified and implemented new incremental learning algorithms for DFA and Kripke structures.These have been integrated with the NuSMV model checker to give a new learning-based testing architecture. We have evaluated our architecture on case studies and shown that the method is effective. / QC 20110822

In a broader sense the main research objective of this thesis (and ongoing research work) is distributed knowledge management for mobile dynamic systems. But the primary focus and presented work focuses on communication/interoperability of heterogeneous entities in an infrastructure less paradigm, a distributed resource manipulation infrastructure and distributed learning in the absence of global knowledge. The research objectives achieved discover the design aspects of heterogeneous distributed knowledge systems towards establishing a seamless integration. This thesis doesn’t cover all aspects in this work; rather focuses on interoperability and distributed learning. Firstly a discussion on the issues in knowledge management for swarm of heterogeneous entities is presented. This is done in a broader and rather abstract fashion to provide an insight of motivation for interoperability and distributed learning towards knowledge management. Moreover this will also serve the reader to understand the ongoing work and research activities in much broader perspective. Primary focus of this thesis is communication/interoperability of heterogeneous entities in an infrastructure less paradigm, a distributed resource manipulation infrastructure and distributed learning in the absence of global knowledge. In dynamic environments for mobile autonomous systems such as robot swarms or mobile software agents there is a need for autonomic publishing and discovery of resources and just-in-time integration for on-the-fly service consumption without any a priori knowledge. SOA (Service-Oriented Architecture) serves the purpose of resource reuse and sharing of services different entities. Web services (a SOA manifestation) achieves these objectives but its exploitation in dynamic environments, where the communication infrastructure is lacking, requires a considerable research. Generally Web services are exploited in stable client-server paradigms, which is a pressing assumption when dynamic distributed systems are considered. UDDI (Universal Description Discovery and Integration) is the main pediment in the exploitation of Web services in distributed control and dynamic natured systems. UDDI can be considered as a directory for publication and discovery of categorized Web services but assumes a centralized registry; even if distributed registries and associated mechanism are employed problems of collaborative communication in infrastructure less paradigms are ignored. Towards interoperability main contribution this thesis is a mediator-based distributed Web services discovery and invocation middleware, which provides a collaborative and decentralized services discovery and management middleware for infrastructure-less mobile dynamic systems with heterogeneous communication capabilities. Heterogeneity of communication capabilities is abstracted in middleware by a conceptual classification of computing entities on the basis of their communication capabilities and communication issues are resolved via conceptual overlay formation for query propagation in system. The proposed and developed middleware has not only been evaluated extensively using Player Stage simulator but also been applied in physical robot swarms. Experimental validations analyze the results in different communication modes i. active and ii. passive mode of communication with and without shared resource conflict resolution. I analyze discoverable Web services with respect to time, services available in complete view of cluster and the impact and resultant improvements in distributed Web services discovery by using caching and semantics. Second part of this thesis focuses on distributed learning in the absence of global information. This thesis takes the argument of defeasibility (common-sense inference) as the basis of intelligence in human-beings, in which conclusions/inferences are drawn and refuted at the same time as more information becomes available. The ability of common-sense reasoning to adapt to dynamic environments and reasoning with uncertainty in the absence of global information seems to be best fit for distributed learning for dynamic systems. This thesis, thus, overviews epistemic cognition in human beings, which motivates the need of a similar epistemic cognitive solution in fabricated systems and considers formal concept analysis as a case for incremental and distributed learning of formal concepts. Thesis also presents a representational schema for underlying logic formalism and formal concepts. An algorithm for incremental learning and its use-case for robotic navigation, in which robots incrementally learn formal concepts and perform common-sense reasoning for their intelligent navigation, is also presented. Moreover elaboration of the logic formalism employed and details of implementation of developed defeasible reasoning engine is given in the latter half of this thesis. In summary, the research results and achievements described in this thesis focus on interoperability and distributed learning for heterogeneous distributed knowledge systems which contributes towards establishing a seamless integration in mobile dynamic systems. / QC 20100614 / ROBOSWARM EU FP6

The combination of machine learning on embedded systems has quickly increased throughout the years. Subsets like TinyML have become an integral part of how embedded systems implement machine learning. The field has evolved quickly, and TinyOL is an emerging subset that redefines what is possible with embedded systems. This report presents a comparison of how a neural network that implements incremental online learning learns and adapts how to do simple tasks in home automation. The comparison is done with another system, mainly a proportional-integral-derivative (PID). The systems were tasked with controlling an LED lightning threshold based on feedback from the user. The systems were evaluated based on their mean absolute error (MAE) and accuracy in predicting the output of the LED lighting system. The MAE values of both systems were compared for different target outputs and threshold values, and the accuracy was calculated by comparing the number of successful iterations to the total number of iterations. The results show that the neural network has an accuracy of 50\% when a learning rate of 0.2 is used, 97.5\% when a learning rate of 0.5 is used, and 47.5\% when a learning rate of 1.0 is used. The PID control system had accuracy values of 45\% when using an adaption rate of 0.2, 47.5\% when using an adaption rate of 0.5, and 90\% when using an adaption rate of 1.0. The neural network also showcased a lower median MAE for every test conducted. The study provides insights into the effectiveness of different control systems and can inform the development of similar systems in the future. / Kombinationen av maskininlärning på inbygga system har snabbt ökat under åren. Tekniker som TinyML har snabbt blivit en integrerad del av hur inbyggda system implementerar maskininlärning. Teknikerna har snabbt utvecklats och TinyOL är en framväxande delmängd av TinyML som omdefinierar vad som är möjligt med inbyggda system. Denna rapport presenterar en jämförelse av hur ett neuralt nätverk som implementerar inkrementell online-inlärning lär och anpassar sig för att utföra enklare uppgifter inom hemautomation. Jämförelsen görs med ett annat system, huvudsakligen en proportional-integral-derivative (PID). Systemen fick i uppgift att kontrollera en LED-lampa baserat på användarens feedback. Systemet utvärderas baserat på deras mean absolute error (MAE) och noggrannhet i att förutsäga börvärdet för LED-belysningen. MAE-värderna för båda systemen jämfördes för de olika målen och börvärdena, och noggrannheten beräknades genom att jämföra antalet lyckade iterationer med det totala antalet iterationer. Resultaten visar att neurala nätverket har en  noggrannhet på  50\% när en learning rate på 0.2 användes, 97.5\% när en learning rate på 0.5 användes och 47.5\% när en learning rate på 1.0 användes. PID kontroll system hade en noggranhet på 45\% när en adaption rate på 0.2 användes, 47.5\% när en adaption rate på 0.5 användes och 90\% när en adaption rate på 1.0 användes. Det neurala nätverket visade också ett lägre MAE-värde på alla de testerna som utfördes. Studien ger insikter i effektiviteten hos olika kontrollsystem och kan hjälpa utvecklingen av liknande system i framtiden.