Multi-relational learning has become popular due to the limitations of propositional problem definition in structured domains and the tendency of storing data in relational databases. As patterns involve multiple relations, the search space of possible hypotheses becomes intractably complex. Many relational knowledge discovery systems have been developed employing various search strategies, search heuristics and pattern language limitations in order to cope with the complexity of hypothesis space. In this work, we propose a relational concept learning technique, which adopts concept descriptions as associations between the concept and the preconditions to this concept and employs a relational upgrade of association rule mining search heuristic, APRIORI rule, to effectively prune the search space. The proposed system is a hybrid predictive inductive logic system, which utilizes inverse resolution for generalization of concept instances in the presence of background knowledge and refines these general patterns into frequent and strong concept definitions with a modified APRIORI-based specialization operator. Two versions of the system are tested for three real-world learning problems: learning a linearly recursive relation, predicting carcinogenicity of molecules within Predictive Toxicology Evaluation (PTE) challenge and mesh design. Results of the experiments show that the proposed hybrid method is competitive with state-of-the-art systems.

Made available in DSpace on 2016-06-02T20:30:55Z (GMT). No. of bitstreams: 1 5022.pdf: 1889377 bytes, checksum: f804286a43c335f751c6d1742038211b (MD5) Previous issue date: 2013-02-25 / Universidade Federal de Minas Gerais / Exclusion responding has been considered one of the processes by which children learn to relate new words to objects. This pattern of responding is usually tested with a matching to sample task. The goal of the present experiment was to verify if exclusion responding also occurs with properties-object relations: dictated model stimuli were adjectives (/happy/, /sad/ and /anger/) and comparison stimuli were properties of objects (dolls facial expression). A second goal was to investigate the relation between the quantity of exclusion trials and a learning outcome. Phase 1 established a baseline of auditory-visual discriminations in a natural situation with dolls; Phase 2 established the baseline among dictated words and pictures of the doll (a girl) with a specific software presented with a notebook; Phase 3 consisted of exclusion probes in which the pseudo words /piva/ and /fobam/ were presented as models in different trials and a defined face (related to baseline words), an undefined face (not related to baseline words) and a mask (blank comparison) were presented as comparison. Selection of undefined face showed exclusion control. Phase 3 also consisted of presentation of learning probes. In Phase 4 experimental blocks presented as comparison stimuli figures of another doll (boy) expressing the same emotions of the doll of Phase 3. The purpose of Phase 4 was to evaluate generalization of the results obtained in Phase 3. Experiment was finished if the participant had responded correctly the three learning probes or if he or she had been exposed to 10 exclusion trials with each relation. Data was collected with 22 typically developing preschool children between 24 and 29 months of age. All participants responded by exclusion along the experiment. 11 of the 22 participants showed a learning outcome, responding correctly the three learning probes (100% of correct responses) with one of the taught relations in Phase 3 or with one of the tested relations in Phase 4. Data confirm and extend regularity of exclusion responding. The amount of exclusion trials necessary to obtain a learning outcome and the best learning probes to be used still needs to be investigated. / O responder por exclusão é considerado um dos processos pelos quais as crianças aprendem a relacionar palavras novas a objetos. A tarefa experimental típica é a de escolha de acordo com o modelo. Este estudo teve como objetivo verificar se o responder por exclusão ocorre no ensino de relações palavra-qualidade (adjetivos): os estímulos modelo ditados eram adjetivos (/feliz/, /triste/, /brava/) e os estímulos de comparação visuais eram faces estilizadas de um fantoche. Verificou-se também a relação entre a quantidade de tentativas de exclusão e a aprendizagem de novas relações palavra-qualidade. A Fase 1 consistiu no estabelecimento da linha de base de discriminações auditivo-visuais em uma situação natural. Na Fase 2, a linha de base foi estabelecida em um computador portátil. A Fase 3 consistiu na apresentação de tentativas de exclusão em que duas pseudopalavras /piva/ e /fobam/ eram ditadas como modelos em diferentes tentativas. Os estímulos de comparação eram uma face definida (relacionada à linha de base), uma face indefinida (não relacionada à linha de base) e um estímulo comparação vazio (máscara). A seleção da face indefinida evidenciaria o controle por exclusão. Durante a Fase 3 foram apresentadas também sondas de aprendizagem. Na Fase 4, os blocos experimentais apresentavam como estímulos de comparação bonecos do sexo masculino com as mesmas expressões faciais das do fantoche da Fase 3. Essa fase teve como objetivo verificar a generalização do padrão comportamental dos participantes na fase anterior. O procedimento era finalizado caso o participante acertasse as três sondas de aprendizagem ou após o máximo de 10 tentativas de exclusão com cada relação. Os participantes, 22 crianças pré-escolares, com desenvolvimento típico e idades entre 24 e 29 meses, responderam por exclusão em praticamente todas as sondas ao longo do experimento. Desses, 11 apresentaram desempenho correspondente com aprendizagem em pelo menos uma das novas relações ensinadas na Fase 3 e avaliadas na Fase 4. Os dados confirmam e estendem a regularidade do responder por exclusão. A quantidade de exposição a tentativas de exclusão necessárias para assegurar aprendizagem e a validade dos tipos de sondas de aprendizagem ainda devem ser verificadas experimentalmente.

Made available in DSpace on 2016-06-02T20:30:56Z (GMT). No. of bitstreams: 1 5453.pdf: 1352961 bytes, checksum: 6c7a8392ec5e53e805f81c028682190a (MD5) Previous issue date: 2012-07-30 / Universidade Federal de Minas Gerais / The investigation of symbolic or pre-symbolic behavior in dogs requires the establishment of arbitrary relations among stimuli. A way to teach such relations is requiring a common response in the presence of each one of them, which can result in a class of functionally equivalent stimuli. This study aimed to investigate the formation of functional classes with four dogs. For this purpose it was used an automated device for emission and recording of operant responses and for the presentation of visual stimuli. The operant response was to nose poke a stimuli which were presented in a touch-screen monitor. Three sets of two stimuli (A1/A2; B1/B2; C1/C2) were used for three dogs and two sets of two stimuli (C1/C2; E1/E2) were used for the other dog. In each set, one stimulus was related to reinforcement (S+) and the other to extinction (S-), regarding tasks of simultaneous simple discrimination. Five experimental phases were programmed for three dogs: I) Training and reversals of pair A stimuli; II) Training and reversals of pair B; III) Training and reversions of AB being presented in the same session with functional class formation probes; IV) Training and reversals of pair C; and V) Training and reversals of ABC with functional class formation probes. On the third and fifth phases stimuli of the same set (A1 and B1 or A1, B1 and C1, respectively) were established as S+, and the stimuli of the other set (A2 and B2 or A2, B2 and C2) as S-. After baseline was acquired and reached stability the functions of the stimuli were reversed repeatedly (S+ began to function as S- and vice versa), and it was appraised if the animals reverted the functions of the other stimuli given the reversal of the first pair in one of the sets, and that before direct exposure to new contingencies which would indicate functional classes formation (S+ class and S- class). For the fourth dog four experimental phases were programmed: I) Training of pair E stimuli (without reversal); II) Training of pair C stimuli; III) Training of CE being presented in the same session; and IV) Three reversions of CE being presented in the same session. The results show that the procedure was able to establish a complex and flexible discriminative repertoire in dogs, but insufficient to demonstrate relational responding in the only animal to be exposed to the functional class probes. Nevertheless, considerations were made about the positive aspects of the proposed procedure and learning set formation. / A investigação de comportamentos simbólicos ou pré-simbólicos em cães requer o estabelecimento de relações arbitrárias entre estímulos. Uma alternativa para ensinar tais relações é a exigência de uma resposta comum na presença de cada um deles, o que pode resultar em uma classe de estímulos funcionalmente equivalentes. O presente estudo teve por objetivo investigar a formação de classes funcionais em quatro cães. Para esta finalidade foi utilizado um equipamento automático para a emissão e registro das respostas operantes e apresentação dos estímulos visuais. A resposta operante era tocar com o focinho sobre estímulos projetados em um monitor equipado com tela sensível ao toque. Para três cães foram empregados três conjuntos de dois estímulos (A1/A2; B1/B2; C1/C2) e para o outro cão dois conjuntos de dois estímulos (C1/C2; E1/E2). Em cada conjunto um dos estímulos do par era relacionado com reforço (S+) e o outro era relacionado com extinção (S-), em tarefas de discriminação simples simultânea. Para os três primeiros cães foram programadas cinco fases experimentais: I) Treino e reversões com estímulos do par A; II) Treino e reversões com o par B; III) Treino e reversões de AB, apresentados em uma mesma sessão, com sondas de formação de classes funcionais; IV) Treino e reversões do par C e V) Treino e reversões de ABC, com apresentação de sondas de formação de classes funcionais. Na terceira e quinta fases os estímulos de um mesmo conjunto (A1 e B1 ou A1, B1 e C1, respectivamente) eram estabelecidos como S+ e os estímulos do outro conjunto (A2 e B2 ou A2, B2 e C2) como S-. Após aquisição e estabilidade da linha de base, as funções dos estímulos eram revertidas (os S+ passavam a funcionar como S- e vice-versa) repetidas vezes e era avaliado se, a partir da reversão do primeiro par de estímulos de um dos conjuntos, os animais revertiam as funções dos outros estímulos, antes da exposição direta às novas contingências, evidenciando formação de classes funcionais (a classe dos S+ e a classe dos S-). Para o quarto animal foram programadas quatro fases experimentais: I) Treino do par E (sem reversões) II) Treino do par C III) Treino dos pares C e E apresentados em uma mesma sessão e IV) Três reversão sucessivas dos pares C e E apresentados na mesma sessão. Os resultados obtidos mostram que o procedimento adotado foi capaz de estabelecer um repertório discriminativo complexo e flexível em cães, porém insuficiente para demonstrar responder relacional no único animal a passar pelas sondas de formação de classe funcional. Apesar disso, foram feitas considerações a respeito de aspectos positivos sobre o procedimento proposto e a formação de learning set.

Os idosos representam um grupo crescente na população e algumas alterações importantes de saúde dessa população advêm de acidentes vasculares cerebrais, que podem deixar sequelas como as afasias. A despeito de resultados positivos obtidos por áreas tradicionais do tratamento das afasias, analistas do comportamento têm desenvolvido procedimentos de ensino bem-sucedidos com populações clínicas de não idosos, empregando procedimentos de exclusão e de ensino de repertórios de discriminação condicional para formação de classes de equivalência entre estímulos verbais. Os objetivos gerais deste trabalho foram: a) verificar a efetividade de um procedimento de ensino por exclusão de substantivos e de verbos para afásicos fluentes e não fluentes, a partir do ensino de discriminações condicionais com estímulos auditivos e visuais, e b) verificar a emergência de repertórios não diretamente ensinados (relações transitivas entre estímulos, nomeação e leitura). Participaram do estudo 14 idosos com afasia. Foram conduzidos dois estudos, nos quais os participantes foram expostos às avaliações iniciais de repertório verbal e foram submetidos ao procedimento de ensino de relações entre palavras ditadas, figuras ou vídeos, e palavras impressas, com sondas posteriores de formação de equivalência, nomeação e leitura, além de pós-testes. Foi utilizado um delineamento de múltiplas sondagens. Resultados gerais indicaram que o procedimento foi efetivo para ensinar novas relações aos participantes, porém não foi suficiente para sustentar a formação de classes de equivalência entre estímulos para a maioria das relações ensinadas, e nem para a emergência de nomeação desses estímulos. Não houve diferenças entre a aprendizagem de substantivos e verbos. O procedimento estudado, com adaptações, pode ser uma alternativa terapêutica para recuperação de repertórios verbais em afásicos. / The elderly represents a growing group in the population and some important changes in the health of this population come from strokes, which can leave sequels such as aphasias. Despite positive results from traditional areas of aphasia treatment, behavioral analysts have developed successful teaching procedures with non-elderly clinical populations, employing exclusion procedures and teaching conditional discrimination to form equivalence classes between verbal stimuli. The general objectives were: a) to verify the effectiveness of a teaching by exclusion nouns and verbs for fluent and non-fluent aphasics, from teaching conditional discriminations with auditory and visual stimuli, and b) verify the emergence of repertoires not directly taught (transitive relations between stimuli, naming and reading). A total of 14 elderly people with aphasia participated in the study. Two studies were conducted, in which the participants were exposed to the initial verbal repertoire evaluations and were submitted to the procedure of teaching relationships between dictated words, figures or videos, and printed words, with subsequent probes of formation of equivalence, naming and reading, besides post-tests. A multiple probe design was used. Overall results indicate that the procedure was effective to teach new relationships to the participants, but it was not enough to sustain the formation of equivalence classes between stimuli for most of the relationships taught, and the emergence of naming these stimuli. There were no differences between the learning of nouns and verbs. The procedure studied, with adaptations, may be a therapeutic alternative for the recovery of verbal repertoires in aphasics.

Relational data is characterised by the rich structure it encodes in the dependencies between the individual entities of a given domain. Statistical Relational Learning (SRL) combines first-order logic and probability to learn and reason over relational domains by creating parametric probability distributions over relational structures. SRL models can succinctly represent the complex dependencies in relational data and admit learning and inference under uncertainty. However, these models are significantly limited when it comes to the tractability of learning and inference. This limitation emerges from the intractability of Weighted First Order Model Counting (WFOMC), as both learning and inference in SRL models can be reduced to instances of WFOMC. Hence, fragments of first-order logic that admit tractable WFOMC, widely known as domain-liftable, can significantly advance the practicality and efficiency of SRL models. Recent works have uncovered another limitation of SRL models, i.e., they lead to unintuitive behaviours when used across varying domain sizes, violating fundamental consistency conditions expected of sound probabilistic models. Such inconsistencies also mean that conventional machine learning techniques, like training with batched data, cannot be soundly used for SRL models. In this thesis, we contribute to both the tractability and consistency of probabilistic inference in SRL models. We first expand the class of domain-liftable fragments with counting quantifiers and cardinality constraints. Unlike the algorithmic approaches proposed in the literature, we present a uniform combinatorial approach, admitting analytical combinatorial formulas for WFOMC. Our approach motivates a new family of weight functions allowing us to express a larger class of probability distributions without losing domain-liftability. We further expand the class of domain-liftable fragments with constraints inexpressible in first-order logic, namely acyclicity and connectivity constraints. Finally, we present a complete characterization for a statistically consistent (a.k.a projective) models in the two-variable fragment of a widely used class of SRL models, namely Markov Logic Networks.

No description available.

Made available in DSpace on 2016-06-02T19:44:13Z (GMT). No. of bitstreams: 1 4832.pdf: 5480712 bytes, checksum: f883b2966beb312d72c283bb6a1722d8 (MD5) Previous issue date: 2012-12-08 / Financiadora de Estudos e Projetos / The overall aim of this thesis was to investigate relevant variables to relational learning and the emergence of symbolic behavior in people with autism, in the teaching of reading. Were conducted three studies that are independent and yet complementary. Chapter 1 describes a study that evaluated the performance of forty participants with autism and forty with intellectual disabilities in identity matching-to-sample task, in the typical organization (one model and three comparisons) and in the organization with more than one model (three models and three comparisons), in two forms of presentation that needed different response modes: on tabletop and by computer. The general results of the participants indicated better performance in the organization with more than one model, in paper and in computer. Chapter 2 presents a study that aimed to teach basic reading skills to five participants with autism (four speakers and one nonspeaker), from the teaching of arbitrary relations among pictures and printed words, using adapted matching-to-sample task. Participants learned basic reading skills, but not showed combinatorial reading. Chapter 3 presents a study designed to teach combinatorial reading with comprehension for three participants with autism, from the teaching of naming syllables, words, pictures, and the formation of stimulus equivalence classes. Participants learned combinatorial reading with comprehension, with few training sessions and low numbers of errors during the teaching. The set of studies indicate the possibility of symbolic learning by people with autism since appropriate teaching procedures are used. / O objetivo geral dessa tese foi investigar variáveis relevantes para a aprendizagem relacional e para a emergência de comportamento simbólico em pessoas com autismo, no âmbito do ensino de leitura. Foram realizados três estudos que são independes e ao mesmo tempo complementares. O Capitulo 1 descreve um estudo que avaliou o desempenho de quarenta participantes com autismo e quarenta com deficiência intelectual em tentativas de emparelhamento com modelo por identidade, na organização típica (um modelo e três comparações) e na multimodelo (três modelos e três comparações), em duas formas de apresentação que requeriam modos de resposta diferentes: em papel e por computador. Os dados gerais dos participantes indicaram melhores desempenhos nas tentativas multimodelo, em papel e no computador. O Capitulo 2 apresenta um estudo que pretendeu ensinar habilidades básicas de leitura a cinco participantes com autismo (quatro falantes e um não falante), a partir do ensino de relações arbitrárias entre figuras e palavras impressas, por meio de tentativas de emparelhamento com o modelo multimodelo. Os participantes aprenderam habilidades básicas e iniciais de leitura, porém não demonstraram leitura recombinativa com compreensão. O Capítulo 3 apresenta um estudo planejado para ensinar leitura combinatória com compreensão a três participantes com autismo, a partir do ensino da nomeação de sílabas, palavras, figuras e da formação de classes de estímulos equivalentes. Os participantes aprenderam leitura combinatória com compreensão, com poucas sessões de treino e com baixo número de erros durante o ensino. O conjunto de estudos indica a possibilidade da aprendizagem simbólica por pessoas com autismo desde que procedimentos de ensino adequados sejam utilizados.

Made available in DSpace on 2016-08-17T14:53:05Z (GMT). No. of bitstreams: 1 Lucas Rego Drumond.pdf: 6150160 bytes, checksum: 27ad4ea0ffdf273a78782ada8f04da6b (MD5) Previous issue date: 2009-10-23 / Knowledge representation formalisms, such as ontologies, have proven to be a powerful tool for enhancing the effectiveness of natural language processing, information filtering and retrieval and so on. Besides these tasks, ontologies are also crucial for the Semantic Web, a new generation of the Web that aims at structuring its content in such a way that it can be more effectively processed by machines. However, knowledge systems suffer from the so called knowledge acquisition bottleneck, i.e. the difficulty in constructing knowledge bases. An approach for this problem is to provide automatic or semi-automatic support for ontology construction, a field of research known as ontology learning. This work discusses the state of the art of ontology learning techniques and proposes and approach for supporting the ontology construction process through the automatization of the concept hierarchy extraction from textual sources. The proposed process is composed by two techniques, namely PRECE (Probabilistic Relational Concept Extraction) and PREHE (Probabilistic Relational Hierarchy Extraction). The PRECE technique extracts ontology concepts from textual sources while the PREHE technique extracts taxonomic relationships between the concepts extracted by PRECE. Both techniques use Markov logic networks, an approach for statistical relational learning that combines first order logic with Markov networks. The PRECE and PREHE techniques were evaluated in the touristic domain and their results were compared with an ontology manually developed by a domain expert. / Os formalismos de representação do conhecimento como as ontologias têm se mostrado uma poderosa ferramenta para melhorar a efetividade de sistemas de processamento da linguagem natural, recuperação e filtragem de informação e muitas outras tarefas. Além disso, as ontologias são essenciais para a Web Semântica, uma nova geração da Web que visa estruturar o conteúdo da mesma de modo que este possa ser processado de forma mais efetiva pelas máquinas. Entretanto, os sistemas de conhecimento sofrem do problema conhecido como o gargalo da aquisição do conhecimento, que nada mais é do que a dificuldade de construção das bases de conhecimento. Uma abordagem para este problema é o suporte automático ou semi-automático à construção de ontologias. Este campo de pesquisa é conhecido como aprendizagem de ontologias. Este trabalho discute o estado da arte das técnicas de aprendizagem de ontologias e propõe uma abordagem para o suporte ao processo de construção de ontologias através da automatização da extração de hierarquias de conceitos a partir de fontes textuais. O processo proposto é composto por duas técnicas, a PRECE (Probabilistic Relational Concept Extraction) para a extração de conceitos e a PREHE (Probabilistic Relational Hierarchy Extraction) para a descoberta de relacionamentos taxonômicos entre os conceitos extraídos pela PRECE. As duas técnicas fazem uso das Redes Lógicas de Markov, uma abordagem da aprendizagem probabilística relacional que combina a lógica de primeira ordem com as redes de Markov. As técnicas PRECE e PREHE foram avaliadas no domínio turístico comparando os seus resultados com uma ontologia desenvolvida manualmente por especialistas neste domínio.

<p>Language “in the wild” is complex and ambiguous and relies on a shared understanding of the world for its interpretation. Most current natural language processing methods represent language by learning word co-occurrence patterns from massive amounts of linguistic data. This representation can be very powerful, but it is insufficient to capture the meaning behind written and spoken communication. </p> <p> </p> <p>In this dissertation, I will motivate neural-symbolic representations for dealing with these challenges. On the one hand, symbols have inherent explanatory power, and they can help us express contextual knowledge and enforce consistency across different decisions. On the other hand, neural networks allow us to learn expressive distributed representations and make sense of large amounts of linguistic data. I will introduce a holistic framework that covers all stages of the neural-symbolic pipeline: modeling, learning, inference, and its application for diverse discourse scenarios, such as analyzing online discussions, mining argumentative structures, and understanding public discourse at scale. I will show the advantages of neural-symbolic representations with respect to end-to-end neural approaches and traditional statistical relational learning methods.</p> <p><br></p> <p>In addition to this, I will demonstrate the advantages of neural-symbolic representations for learning in low-supervision settings, as well as their capabilities to decompose and explain high-level decision. Lastly, I will explore interactive protocols to help human experts in making sense of large repositories of textual data, and leverage neural-symbolic representations as the interface to inject expert human knowledge in the process of partitioning, classifying and organizing large language resources. </p>

<p>Nowadays an increasing number of people consume, share, and interact with information online. This results in posting and counter-posting on online media by different ideological groups on various polarized topics. Consequently, online media has become the primary platform for political and social influencers to directly interact with the citizens and share their perspectives, views, and stances with the goal of gaining support for their actions, bills, and legislation. Hence, understanding the perspectives and the influencing strategies in online media texts is important for an individual to avoid misinformation and improve trust between the general people and the influencers and the authoritative figures such as the government.</p> <p><br></p> <p>Automatically understanding the perspectives in online media is difficult because of two major challenges. Firstly, the proper grammar or mechanism to characterize the perspectives is not available. Recent studies in Natural Language Processing (NLP) have leveraged resources from social science to explain perspectives. For example, Policy Framing and Moral Foundation Theory are used for understanding how issues are framed and the moral appeal expressed in texts to gain support. However, these theories often fail to capture the nuances in perspectives and cannot generalize over all topics and events. Our research in this dissertation is one of the first studies that adapt social science theories in Natural Language Processing for understanding perspectives to the extent that they can capture differences in ideologies or stances. The second key challenge in understanding perspectives in online media texts is that annotated data is difficult to obtain to build automatic methods to detect the perspectives, that can generalize over the large corpus of online media text on different topics. To tackle this problem, in this dissertation, we used weak sources of supervision such as social network interaction of users who produce and interact with the messages, weak human interaction, or artificial few-shot data using Large Language Models. </p> <p><br></p> <p>Our insight is that various tasks such as perspectives, stances, sentiments toward entities, etc. are interdependent when characterizing online media messages. As a result, we proposed approaches that jointly model various interdependent problems such as perspectives, stances, sentiments toward entities, etc., and perform structured prediction to solve them jointly. Our research findings showed that the messaging choices and perspectives on online media in response to various real-life events and their prominence and contrast in different ideological camps can be efficiently captured using our developed methods.</p>