Denna uppsats presenterar forskning som belyser de förändringar inom utbildningsystemen somföljt i och med utvecklingen av internet, sociala medier, online kommunikation och utbyte avkunskap online. Under det senaste årtiondet har sättet människor lär sig på förändrats från dettraditionella klassrummet, som enbart använder sig av tryckt material, till det modernaklassrummet som nyttjar digitalt material [1]. Electronic learning (e-learning) innebär av attanvända sig av digitala material istället för det traditionella tryckt materialet [2]. Mobilelearning tar detta koncept ett steg längre genom att kombinera individuellt lärande medvarsomhelst och närsomhelst lärande [2]. Detta möjliggörs till stor del av den teknologiskautvecklingen inom mobila enheter [3]. Relational learning är ett sätt att lära där individen lärfrån andra genom ett gemensamt utbyte av idéer och kunskap [4].Parkour, eller konsten av rörelse, har av O’Grady blivit presenterat som en kollaborativläromodell [5].Utövare av parkour kallas traceurs och när traceurs utövar parkour blir parkourett verktyg för aktivt lärande. Med detta lärande blir förståelse och överkommandet av hinderen läromodell. Detta är en läromodell där vetandet och upprepandet är nyckeln till traceurssträvan efter lärandets berusning. Med lek förvandlar parkour sin omgivning till en miljö förlärande. Med hjälp av parkour tar traceurs över ägandet av sitt egna lärande och hittarmöjligheter att kunna uttrycka sig själva [6]. En naturlig del av parkour är vikten som läggs påatt vara medveten om sina egna förmågor och mål med sitt lärande. Genom att dela med sigav resultatet av detta lärande skapas basen av den kollaborativa läromiljön hos parkour.I denna uppsats presenteras en mobil Android applikation: Vault. Vault kombinerar mobilelearning med relational learning och använder sig av fördelarna av den kollaborativaläromodellen som existerar inom parkour. Tre populära mobila applikationer ämnade för lärandehar analyserats. Denna analys i samband resterande resultat har varit med och format designenoch utveckling av den presenterade applikationen Vault.Vault är även testad för att mäta dess potentiella fördelar av att användas som en allmänutbildningsapplikation, med ett fokus på relational learning. Detta test är beskrivet ochresultaten är presenterade. Avslutningsvis återfinns en diskussion angående resultaten från bådetestet av applikationen samt analysen och det teoretiska materialet som presenterats iuppsatsen. Denna diskussion följs av en sammanfattning som även innehåller förslag påframtida forskningsområden som kan utforskas vidare. / This thesis presents research that addresses the educational change that arises in the era ofinternet, social media, online communities, and knowledge sharing on the web. During the lastdecade, the way people learn has seen a big shift from the traditional classroom that purelyuses printed material to the contemporary classroom that utilizes digital technologies for theteaching material [1]. Electronic learning is teaching using electronic resources instead of thetraditional printed material [2]. Mobile learning takes this concept one step further bycombining individualized learning with anytime and anywhere learning [2], enabled by thetechnological advances of mobile devices [3]. Relational learning is a way of learning in whichthe individuals involved learn from each other through the bilateral exchange of experiences andideas [4].Parkour, or the art of movement, has been presented as a collaborative learning model byO’Grady [5]. As parkour practitioner, also known as traceurs, play parkour they also provide aplatform for active learning where knowing and overcoming obstacles composes an educationalmodel. In this model, knowing and repeating is the key to the pursuit of learning. With play,parkour appropriates the spaces in which it takes place into an environment of learning. Byplaying, traceurs take ownership of their own learning process, finding the flow path that letsthem express themselves [6]. Inherent to the practice of parkour is the importance of selfawareness of one’s skills and learning goals, as well as recording and sharing the learningoutcomes. This, in turn, becomes the basic construct of a collaborative learning environment.In this thesis, an Android mobile application, called Vault is presented. Vault combines mobilelearning and relational learning, while at the same time reaping the reward of the communitybased learning model existing in parkour. The thesis also provides an analysis of popular mobilelearning apps. This analysis aides in shaping the design and development of the presentedapplication, Vault.Vault is also tested in order to gauge the potential benefits of using an application designed tobe a general-purpose educational application with a focus on relational learning. This test isdetailed, and the results are presented. The findings from these results, and the results fromthe aforementioned analysis and the theory presented in this thesis, are discussed and futurelines of research are presented.

In this dissertation, I explored relational learning via latent variable models. Traditional machine learning algorithms cannot handle many learning problems where there is a need for modeling both relations and noise. Statistical relational learning approaches emerged to handle these applications by incorporating both relations and uncertainties in these problems. Latent variable models are one of the successful approaches for statistical relational learning. These models assume a latent variable for each entity and then the probability distribution over relationships between entities is modeled via a function over latent variables. One important example of relational learning via latent variables is text data modeling. In text data modeling, we are interested in modeling the relationship between words and documents. Latent variable models learn this data by assuming a latent variable for each word and document. The co-occurrence value is defined as a function of these random variables. For modeling co-occurrence data in general (and text data in particular), we proposed latent logistic allocation (LLA). LLA outperforms the-state-of-the-art model --- latent Dirichlet allocation --- in text data modeling, document categorization and information retrieval. We also proposed query-based visualization which embeds documents relevant to a query in a 2-dimensional space. Additionally, I used latent variable models for other single-relational problems such as collaborative filtering and educational data mining. To move towards multi-relational learning via latent variable models, we propose latent feature networks (LFN). Multi-relational learning approaches model multiple relationships simultaneously. LFN assumes a component for each relationship. Each component is a latent variable model where a latent variable is defined for each entity and the relationship is a function of latent variables. However, if an entity participates in more than one relationship, then it will have a separate random variable for each relationship. We used LFN for modeling two different problems: microarray classification and social network analysis with a side network. In the first application, LFN outperforms support vector machines --- the best propositional model for that application. In the second application, using the side information via LFN can drastically improve the link prediction task in a social network.

A vasta quantidade de informações disponível sobre qualquer área de conhecimento torna cada vez mais difícil selecionar e analisar informações específicas e relevantes sobre determinado assunto. Com isso, faz-se necessário o aprimoramento de técnicas automáticas para recuperação, análise e extração de conhecimento em conjuntos de dados, destacando-se dessa forma as pesquisas em Aprendizado de Máquina e em Mineração de Dados. Em aprendizado de máquina e em mineração, a grande maioria das técnicas utiliza-se de uma representação proposicional dos dados, que considera apenas caracter características individuais dos objetos descritos em uma tabela atributo-valor. Porém, existem aplicações nas quais além da descrição dos objetos também estão disponíveis informações sobre relações existentes entre eles. Esses domínios podem ser representados via grafos, nos quais vértices representam objetos e arestas relações entre objetos, possibilitando a aplicação de técnicas relacionais aos dados. Conceitos de Redes Complexas (RC) podem ser utilizados neste contexto. RC é um campo de pesquisa recente e ativo, que estuda o comportamento de diversos sistemas reais, modelados via grafos. Entretanto, ainda há poucos trabalhos que utilizam Redes Complexas em aprendizado de máquina ou mineração de dados. Este projeto apresenta uma proposta de utilização do formalismo de redes complexas e grafos para descoberta de padrões no contexto de aprendizado supervisionado. O formalismo de grafos permite representar as relações entre objetos e características particulares do domínio, permitindo agregar informações estruturais das relações à descoberta de conhecimento. Especificamente, neste trabalho desenvolve-se uma representação relacional baseada em grafos construídos a partir de relações de similaridade entre objetos. Baseado nesta representação são propostas abordagens de classificação relacional. Também é proposto um modelo de rede denominado K-Associados. Propriedades da rede K-Associados foram investigadas. Os resultados experimentais demonstram um grande potencial para classificação utilizando os algoritmos de classificação e de formação de redes propostos / The vast amount of information available on any area of knowledge makes selecting and analyzing information on a specific topic increasingly dificult. Therefore, it is necessary the improvement of techniques for automatic information retrieval, analysis, and knowledge extraction from data sets. In this scenario, especial attention must be addressed for Machine Learning and Data Mining researches. In machine learning and data mining, most of the techniques uses a propositional representation, which considers only the characteristics of the objects described into an attribute-value table. However, there are domains where, in addition to the description of the objects, it is also available information about relationship between them. Such domains can be represented by graphs where vertices represent objects and edges relationship between objects, enabling the application of techniques for relational data. Concepts of complex networks (CN) can be useful in this context. CN is a recent and active research field, which studies the behavior of many real systems modeled by graphs. However, there is little work in machine learning or data mining applying CN concepts. This project presents a proposal to use the formalism of complex networks and graphs to discover patterns in the context of supervised learning. The formalism of graphs can represent relationships between objects and characteristics of the domain, allowing adding structural knowledge embedded in a graph into the data mining process. Specifically, this work develops a relational representation based on graphs constructed taking into consideration the similarity between objects. Based on this representation, relational classification approaches are proposed. It is also proposed a network referred to K-Associate Network. Properties of the K-Associate Network were investigated. The experimental results show great potential for the proposed classification and network construction algorithms

The use of environmental geometry as a spatial cue is well established for a range of species. Previous research has focused largely on the use of global geometry (e.g., the shape of a room). Thus, comparatively less is known about how local geometry (e.g., corner angles within a room) is encoded. The purpose of the research presented in this thesis was to examine how angular information is encoded and to determine whether angle size influences encoding, using a discrimination task and a spatial array task. Chapter 2 presents a study during which pigeons were trained to discriminate between a small (60°) and large (120°) angle. Once the birds were accurately choosing the angle associated with reward, they were tested on their ability to discriminate between their training angle and one of a series of novel angles. The pigeons showed an absolute learning pattern for the small training angle, but not the large angle. The significance of this result is that the small angle may have been perceived as more distinctive compared to the large angle. Adopting a comparative approach, Chapter 3 presents a study during which adult humans were trained and tested using a similar paradigm but with different training angles (25°, 50° and 75°). The results of this study also support an absolute learning pattern for the small training angle but not the large. These results are significant in that they suggest that angle size may be an important local geometric cue that is encoded in a similar way by both pigeons and humans. To understand how angular information may be processed during a spatial task, Chapter 4 presents a study during which adult humans were trained and tested on their ability to use local angles (either 50° or 75°) to find a goal location within an object array. The results showed that the smaller angle was used more effectively as a spatial cue than the larger angle. Overall, these results are important as they suggest that small and large angles are encoded differently by pigeons and humans, with small angles perceived as more distinctive than large angles.

The use of environmental geometry as a spatial cue is well established for a range of species. Previous research has focused largely on the use of global geometry (e.g., the shape of a room). Thus, comparatively less is known about how local geometry (e.g., corner angles within a room) is encoded. The purpose of the research presented in this thesis was to examine how angular information is encoded and to determine whether angle size influences encoding, using a discrimination task and a spatial array task. Chapter 2 presents a study during which pigeons were trained to discriminate between a small (60°) and large (120°) angle. Once the birds were accurately choosing the angle associated with reward, they were tested on their ability to discriminate between their training angle and one of a series of novel angles. The pigeons showed an absolute learning pattern for the small training angle, but not the large angle. The significance of this result is that the small angle may have been perceived as more distinctive compared to the large angle. Adopting a comparative approach, Chapter 3 presents a study during which adult humans were trained and tested using a similar paradigm but with different training angles (25°, 50° and 75°). The results of this study also support an absolute learning pattern for the small training angle but not the large. These results are significant in that they suggest that angle size may be an important local geometric cue that is encoded in a similar way by both pigeons and humans. To understand how angular information may be processed during a spatial task, Chapter 4 presents a study during which adult humans were trained and tested on their ability to use local angles (either 50° or 75°) to find a goal location within an object array. The results showed that the smaller angle was used more effectively as a spatial cue than the larger angle. Overall, these results are important as they suggest that small and large angles are encoded differently by pigeons and humans, with small angles perceived as more distinctive than large angles.

The ability to recognize sameness among objects and events is a prerequisite for abstraction and forming concepts about what one has learned; thus, identity and nonidentity learning can be considered the backbone of higher-order human cognitive abilities. Discovering identity relations between the constituent properties of objects is an important ability that often characterizes the comparisons that humans make so it is important to devote attention to understanding how nonhuman primates process and conceptualize part-identity as well as whole-identity. Because the ability to generalize the results of learning is to what concepts ultimately reduce, the series of experiments herein first investigated responding to part-identity and -nonidentity and whole-identity and -nonidentity and then explored the generality of such learning to the formation of concepts about color, shape, and cardinal number. The data from Experiments 1, 2, and 3 indicated that the two orangutans learned to respond concurrently to color whole-identity and -nonidentity and they responded faster to color whole-identity. Additionally, both subjects learned to respond concurrently to color and shape part- and whole-identity and for the most part, it was easier for them to do so with color part- and whole-identity problems than shape part- and whole-identity problems. Further, their learned responses to color and shape part- and whole-identity fully transferred to novel color part-identity problems for both subjects and fully transferred to novel color and shape whole-identity problems for one orangutan. The data from Experiments 4, 5, and 6 showed that one subject learned to judge numerical identity when both irrelevant dimensions were cue-constant, but the subject did not do the same when one or more irrelevant dimensions were cue-ambiguous. Further, the subject's accuracy was affected by the numerical distance and the numerical total of comparisons during acquisition of the conditional discrimination. The subject subsequently formed a domain-specific concept about numerical identity as evinced by the transfer of learning to novel numerosities instantiated with novel, cue-constant element colors and shapes and novel numerosities instantiated with cue-constant, familiar element colors and shapes. Given the adaptive significance of using concepts, it is important to investigate if and how nonhuman primates form identity concepts for which they categorize or classify the stimuli around them. This dissertation provided evidence about the extent to which orangutans learned to respond to color, shape, and number identity and nonidentity and subsequent concept formation from such learning. The findings from this study will help in understanding the convergence and divergence in the expression abstraction in the primate phylogeny, thus, informing our understanding about the origins and mechanisms of cognition in human and nonhuman primates.

A vasta quantidade de informações disponível sobre qualquer área de conhecimento torna cada vez mais difícil selecionar e analisar informações específicas e relevantes sobre determinado assunto. Com isso, faz-se necessário o aprimoramento de técnicas automáticas para recuperação, análise e extração de conhecimento em conjuntos de dados, destacando-se dessa forma as pesquisas em Aprendizado de Máquina e em Mineração de Dados. Em aprendizado de máquina e em mineração, a grande maioria das técnicas utiliza-se de uma representação proposicional dos dados, que considera apenas caracter características individuais dos objetos descritos em uma tabela atributo-valor. Porém, existem aplicações nas quais além da descrição dos objetos também estão disponíveis informações sobre relações existentes entre eles. Esses domínios podem ser representados via grafos, nos quais vértices representam objetos e arestas relações entre objetos, possibilitando a aplicação de técnicas relacionais aos dados. Conceitos de Redes Complexas (RC) podem ser utilizados neste contexto. RC é um campo de pesquisa recente e ativo, que estuda o comportamento de diversos sistemas reais, modelados via grafos. Entretanto, ainda há poucos trabalhos que utilizam Redes Complexas em aprendizado de máquina ou mineração de dados. Este projeto apresenta uma proposta de utilização do formalismo de redes complexas e grafos para descoberta de padrões no contexto de aprendizado supervisionado. O formalismo de grafos permite representar as relações entre objetos e características particulares do domínio, permitindo agregar informações estruturais das relações à descoberta de conhecimento. Especificamente, neste trabalho desenvolve-se uma representação relacional baseada em grafos construídos a partir de relações de similaridade entre objetos. Baseado nesta representação são propostas abordagens de classificação relacional. Também é proposto um modelo de rede denominado K-Associados. Propriedades da rede K-Associados foram investigadas. Os resultados experimentais demonstram um grande potencial para classificação utilizando os algoritmos de classificação e de formação de redes propostos / The vast amount of information available on any area of knowledge makes selecting and analyzing information on a specific topic increasingly dificult. Therefore, it is necessary the improvement of techniques for automatic information retrieval, analysis, and knowledge extraction from data sets. In this scenario, especial attention must be addressed for Machine Learning and Data Mining researches. In machine learning and data mining, most of the techniques uses a propositional representation, which considers only the characteristics of the objects described into an attribute-value table. However, there are domains where, in addition to the description of the objects, it is also available information about relationship between them. Such domains can be represented by graphs where vertices represent objects and edges relationship between objects, enabling the application of techniques for relational data. Concepts of complex networks (CN) can be useful in this context. CN is a recent and active research field, which studies the behavior of many real systems modeled by graphs. However, there is little work in machine learning or data mining applying CN concepts. This project presents a proposal to use the formalism of complex networks and graphs to discover patterns in the context of supervised learning. The formalism of graphs can represent relationships between objects and characteristics of the domain, allowing adding structural knowledge embedded in a graph into the data mining process. Specifically, this work develops a relational representation based on graphs constructed taking into consideration the similarity between objects. Based on this representation, relational classification approaches are proposed. It is also proposed a network referred to K-Associate Network. Properties of the K-Associate Network were investigated. The experimental results show great potential for the proposed classification and network construction algorithms

This thesis represents a synthesis of relational learning and causal discovery, two subjects at the frontier of machine learning research. Relational learning investigates algorithms for constructing statistical models of data drawn from of multiple types of interrelated entities, and causal discovery investigates algorithms for constructing causal models from observational data. My work demonstrates that there exists a natural, methodological synergy between these two areas of study, and that despite the sometimes onerous nature of each, their combination (perhaps counterintuitively) can provide advances in the state of the art for both. Traditionally, propositional (or "flat") data representations have dominated the statistical sciences. These representations assume that data consist of independent and identically distributed (iid) entities which can be represented by a single data table. More recently, data scientists have increasingly focused on "relational" data sets that consist of interrelated, heterogeneous entities. However, relational learning and causal discovery are rarely combined. Relational representations are wholly absent from the literature where causality is discussed explicitly. Instead, the literature on causality that uses the framework of graphical models assumes that data are independent and identically distributed. This unexplored topical intersection represents an opportunity for advancement --- by combining relational learning with causal reasoning, we can provide insight into the challenges found in each subject area. By adopting a causal viewpoint, we can clarify the mechanisms that produce previously identified pathologies in relational learning. Analogously, we can utilize relational data to establish and strengthen causal claims in ways that are impossible using only propositional representations.

Learning on relational data is a relevant task in the machine learning community. Extracting information from structured data is a non-trivial task due to the combinatorial complexity of the domain and the necessity to construct methods that work on collections of values of different sizes rather than fixed representations. Relational data can naturally be interpreted as graphs, a class of flexible and expressive structures that can model data from diverse domains,from biology to social interactions. Graphs have been used in a huge variety of contexts, such as molecular modelling, social networks, image processing and recommendation systems. In this manuscript, we tackle some challenges in learning on relational data by developing new learning methodologies. Specifically, in our first contribution, we introduce a new class of metrics for relational data based on relational features extraction technique called Type ExtensionTrees. This class of metrics defines the (dis)similarity of two nodes in a graph by exploiting the nested structure of their relational neighborhood at different depth steps. In our second contribution, we developed a new strategy to collect the information of multisets of data values by introducing a new framework of learnable aggregators called Learning Aggregation Functions.We provide a detailed description of the methodologies and an extensive experimental evaluation on synthetic and real world data to assess the expressiveness of the proposed models. A particular focus is given to the application of these methods to the recommendation systems domain, exploring the combination of the proposed methods with recent techniques developed for Constructive Preference Elicitation and Group Recommendation tasks.

Linked Data (LD) follows the web in providing low barriers to publication, and in deploying web-scale keyword search as a central way of identifying relevant data. As in the web, searchesinitially identify results in broadly the form in which they were published, and the published form may be provided to the user as the result of a search. This will be satisfactory in some cases, but the diversity of publishers means that the results of the search may be obtained from many different sources, and described in many different ways. As such, there seems to bean opportunity to add value to search results by providing userswith an integrated representation that brings together features from different sources. This involves an on-the-fly and automated data integration process being applied to search results, which raises the question as to what technologies might bemost suitable for supporting the integration of LD searchresults. In this thesis we take the view that the problem of integrating LD search results is best approached by assimilating different forms ofevidence that support the integration process. In particular, thisdissertation shows how Statistical Relational Learning (SRL) formalisms (viz., Markov Logic Networks (MLN) and Probabilistic Soft Logic (PSL)) can beexploited to assimilate different sources of evidence in a principledway and to beneficial effect for users. Specifically, in this dissertation weconsider syntactic evidence derived from LD search results and from matching algorithms, semantic evidence derived from LD vocabularies, and user evidence,in the form of feedback. This dissertation makes the following key contributions: (i) a characterisation of key features of LD search results that are relevant to their integration, and a description of some initial experiences in the use of MLN for interpreting search results; (ii)a PSL rule-base that models the uniform assimilation of diverse kinds of evidence;(iii) an empirical evaluation of how the contributed MLN and PSL approaches perform in terms of their ability to infer a structure for integrating LD search results;and (iv) concrete examples of how populating such inferred structures for presentation to the end user is beneficial, as well as guiding the collection of feedbackwhose assimilation further improves search results presentation.