Dissertations / Theses on the topic 'Online continual learning'

La conduite autonome a connu des progrès remarquables au cours des dernières décennies, et la perception machine se présente comme un enjeu fondamental crucial, englobant la détection et le suivi des participants à la circulation tels que les véhicules, les piétons et les cyclistes. Bien que la détection d'objets basée sur la vision ait réalisé des progrès significatifs grâce aux techniques d'apprentissage profond, des défis subsistent dans la détection en 3D.Tout d'abord, les capteurs non visuels, tels que le LiDAR 3D, démontrent des avantages inégalés en matière de détection précise et d'adaptabilité aux conditions d'éclairage variables. Cependant, la complexité liée à la manipulation des données de nuages de points, qui peuvent être difficiles à interpréter, associée au coût élevé de l'annotation manuelle, posent des défis majeurs dans l'utilisation du LiDAR 3D.Deuxièmement, des préoccupations découlent du manque d'interprétabilité des modèles d'apprentissage profond, associée à leur forte dépendance à des données d'entraînement étendues, ce qui nécessite souvent une rétroaction coûteuse pour obtenir une performance acceptable lors de l'adaptation à de nouveaux environnements.Cette thèse aborde ces défis selon trois principaux axes : Génération d'échantillons, Préservation des connaissances et Évitement de l'oubli catastrophique. Nous introduisons le concept d'Apprentissage Continu en Ligne et proposons un cadre général. Ce cadre permet aux modèles de se mettre à jour en temps réel, préservant les connaissances plutôt que les données brutes, et atténuant efficacement la dégradation des performances causée par l'oubli.Les principaux travaux de cette thèse comprennent les trois contributions suivance : 1) Génération d'échantillons : Pour résoudre le problème des nuages de points clairsemés générés par un LiDAR 3D et de l'annotation manuelle laborieuse, nous tirons parti des avantages des données multi-capteurs et utilisons un cadre efficace d'apprentissage en ligne par transfert. Ce cadre transfère efficacement les capacités matures de détection basée sur des images aux détecteurs basés sur le LiDAR 3D. Un aspect innovant est le processus d'apprentissage par utilisation, réalisé grâce à la détection en boucle fermée. Une nouvelle stratégie de fusion d'informations est proposée pour combiner les corrélations spatio-temporelles, améliorant l'efficacité du transfert de connaissances. 2) Préservation des connaissances : L'Apprentissage en Ligne est introduit pour assurer la préservation des connaissances sans conserver les données d'entraînement. Un modèle amélioré de Forêt Aléatoire (Random Forest) en Ligne est intégré, permettant un entrainement rapide du modèle avec des ressources informatiques limitées et un déploiement immédiat. Les paramètres du modèle Forêt Aléatoire en Ligne sont partagés dynamiquement tout au long du processus d’entrainement pour répondre à la distribution inconnue des données. L'exploration des structures d'arbres Forêt Aléatoire en Ligne garantit l'indépendance du processus d’entrainement, renforçant la capacité du modèle à capturer des schémas et des variations complexes. 3) Évitement de l'oubli catastrophique : Pour faire face au problème inévitable de l'oubli dans le cadre d'apprentissage en ligne lors d'un déploiement à long terme, nous proposons le cadre de l'Apprentissage en Ligne à Long-court Terme (ALLCT). L'ALLCT combine de multiples apprenants à court terme basés sur l'apprentissage en ensemble avec un contrôleur à long terme doté d'un mécanisme de décision probabiliste. Ce cadre assure une maintenance efficace des connaissances et s'adapte aux changements lors d'un déploiement à long terme, sans faire d'hypothèses sur les types de modèles et la continuité des données. Les évaluations croisées de jeux de données sur des tâches telles que la détection 3D de participants à la circulation routière démontrent l'efficacité de l'ALLCT dans l'évitement de l'oubli
Autonomous driving has witnessed remarkable progress over the past decades, and machine perception stands as a critical foundational issue, encompassing the detection and tracking of road participants such as vehicles, pedestrians, and cyclists. While vision-based object detection has achieved significant progress thanks to deep learning techniques, challenges still exist in 3D detection.Firstly, non-visual sensors, such as 3D LiDAR, demonstrate unparalleled advantages in achieving precise detection and adaptability to varying lighting conditions. However, the complexity of handling points cloud data, which can be challenging to interpret, coupled with the high cost of manual annotation, pose primary challenges in the use of 3D LiDAR.Secondly, concerns arise from the lack of interpretability in deep learning models, coupled with their heavy reliance on extensive training data, which often necessitates costly retraining for acceptable generalization performance when adapting to new scenes or environments.This dissertation addresses these challenges from three main perspectives: Generation of Samples, Preservation of Knowledge, and Avoidance of Catastrophic Forgetting. We introduce the concept of Online Continual Learning (OCL) and propose a general framework that encompasses detection, tracking, learning, and control. This framework enables models to update in real-time, preserving knowledge rather than raw data, and effectively mitigating the performance degradation caused by catastrophic forgetting.The main work of this dissertation includes: 1) Generation of Samples: To address sparse point clouds generated by 3D LiDAR and the labor-intensive manual annotation, we leverage the advantages of multi-sensor data and employ an efficient online transfer learning framework. This framework effectively transfers mature image-based detection capabilities to 3D LiDAR-based detectors. An innovative aspect is the "learn-by-use" process, achieved through closed-loop detection, facilitating continuous self-supervised learning. A novel information fusion strategy is proposed to combine spatio-temporal correlations, enhancing the effectiveness of knowledge transfer. 2) Preservation of Knowledge: Online Learning (OL) is introduced to address knowledge preservation without retaining training data. An improved Online Random Forest (ORF) model is incorporated, enabling rapid model training with limited computational resources and immediate deployment. The ORF model's parameters are dynamically shared throughout the training process to address the unknown data distribution. The exploration of ORF tree structures ensures independence in training processes, enhancing the model's ability to capture complex patterns and variations. Implementing octrees improves storage efficiency and model access. 3) Avoidance of Catastrophic Forgetting: To tackle the inevitable forgetting problem in online learning frameworks during long-term deployment, we propose the Long Short-Term Online Learning (LSTOL) framework. LSTOL combines multiple short-term learners based on ensemble learning with a long-term controller featuring a probabilistic decision mechanism. This framework ensures effective knowledge maintenance and adapts to changes during long-term deployment, without making assumptions about model types and data continuity. Cross-dataset evaluations on tasks such as 3D detection of road participants demonstrate the effectiveness of LSTOL in avoiding forgetting

Deep learning (DL) is one of the most advanced AI techniques that has gained much attention in the last decade and has also been applied in many successful applications such as market stock prediction, object detection, and face recognition. The rapid advances in computational techniques like Graphic Processing Units (GPU) and Tensor Processing Units (TPU) have made it possible to train large deep learning models to obtain high accuracy surpassing human ability in some tasks, e.g., LipNet [9] achieves 93% of accuracy compared with 52% of human to recognize the word from speaker lips movement. Most of the current deep learning research work has focused on designing a deep architecture working in a static environment where the whole training set is known in advance. However, in many real-world applications like predicting financial markets, autonomous cars, and sensor networks, the data often comes in the form of streams with massive volume, and high velocity has affected the scalability of deep learning models. Learning from such data is called continual, incremental, or online learning. When learning a deep model in dynamic environments where the data come from streams, the modern deep learning models usually suffer the so-called catastrophic forgetting problem, one of the most challenging issues that have not been solved yet. Catastrophic forgetting occurs when a model learns new knowledge, i.e., new objects or classes, but its performance in the previously learned classes reduces significantly. The cause of catastrophic forgetting in the deep learning model has been identified and is related to the weight-sharing property. In detail, the model updating the corresponding weights to capture knowledge of the new tasks may push the learned weights of the past tasks away and cause the model performance to degrade. According to the stability-plasticity dilemma [17], if the model weights are too stable, it will not be able to acquire new knowledge, while a model with high plasticity can have large weight changes leading to significant forgetting of the previously learned patterns. Many approaches have been proposed to tackle this issue, like imposing constraints on weights (regularizations) or rehearsal from experience, but significant research gap still exists. First, current regularization methods often do not simultaneously consider class imbalance and catastrophic forgetting. Moreover, these methods usually require more memory to store previous versions of the model, which sometimes is not able to hold a copy of a substantial deep model due to memory constraints. Second, existing rehearsal approaches pay little attention to selecting and storing critical instances that help the model to retain as much knowledge of the learned tasks. This study focuses on dealing with these challenges by proposing several novel methods. We first proposed a new loss function that combines two loss terms to deal with class imbalance data and catastrophic forgetting simultaneously. The former is a modification of a widely used loss function for class imbalance learning, called Focal loss, to handle the exploding gradient (loss goes to NaN) and the ability to learn from highly confident data points. At the same time, the latter is a novel loss term that addresses the catastrophic forgetting within the current mini-batch. In addition, we also propose an online convolution neural network (OCNN) architecture for tabular data that act as a base classifier in an ensemble system (OECNN). Next, we introduce a rehearsal-based method to prevent catastrophic forgetting. In which we select a triplet of instances within each mini-batch to store in the memory buffer. We find that these instances are identified as crucial instances that can help either remind the model of easy tasks or revise for the hard ones. We also propose a class-wise forgetting detector that monitors the performance of each class encountered so far in a stream. If a class’s performance drops below a predefined threshold, that class is identified as a forgetting class. Finally, based on the nature of data which often comprises many modalities, we study online multi-modal multi-task (M3T) learning problems. Unlike the traditional methods in stable environments, online M3T learning need to be considered in many scenarios like missing modalities and incremental tasks. We establish the setting for six frequently happened scenarios for M3T. Most of the existing works in M3T fail to run on all of these scenarios. Therefore, we propose a novel M3T deep learning model called UniCNet that can work on all of these scenarios and achieves superior performance compared with state-of-the-art M3T methods. To conclude, this dissertation contributes to novel computational techniques that deal with catastrophic forgetting problem in continual deep learning.
Thesis (PhD Doctorate)
Doctor of Philosophy (PhD)
School of Info & Comm Tech
Science, Environment, Engineering and Technology
Full Text

Chat-bots are computer programs designed to conduct textual or audible conversations with a single user. The job of a chat-bot is to be able to find the best response for any request the user issues. The best response is considered to answer the question and contain relevant information while following grammatical and lexical rules. Modern chat-bots often have trouble accomplishing all these tasks. State-of-the-art approaches, such as deep learning, and large datasets help chat-bots tackle this problem better. While there is a number of different approaches that can be applied for different kind of bots, datasets of suitable size are not always available. In this work, we introduce and evaluate a method of expanding the size of datasets. This will allow chat-bots, in combination with a good learning algorithm, to achieve higher precision while handling their tasks. The expansion method uses the continual learning approach that allows the bot to expand its own dataset while holding conversations with its users. In this work we test continual learning with IBM Watson Assistant chat-bot as well as a custom case study chat-bot implementation. We conduct the testing using a smaller and a larger datasets to find out if continual learning stays effective as the dataset size increases. The results show that the more conversations the chat-bot holds, the better it gets at guessing the intent of the user. They also show that continual learning works well for larger and smaller datasets, but the effect depends on the specifics of the chat-bot implementation. While continual learning makes good results better, it also turns bad results into worse ones, thus the chat-bot should be manually calibrated should the precision of the original results, measured before the expansion, decrease.

Dans cette thèse, nous nous intéressons au problème de l'apprentissage continu en ligne des réseaux de neurones artificiels, qui consiste à apprendre en permanence sur un flux de données. La difficulté majeure réside dans le fait qu'intégrer les nouvelles informations présentes dans le flux tend à effacer les connaissances précédemment acquises, un phénomène appelé oubli catastrophique.Dans ce domaine de l'apprentissage continu en ligne, notre recherche porte sur deux applications importantes en vision par ordinateur : la classification d'images et la détection d'objets. Le flux de données est alors constitué d'une suite d'images.En classification d'images le réseau de neurones doit apprendre progressivement à classifier les images des nouvelles classes sans oublier les anciennes. La méthode la plus répandue pour résoudre ce problème est le rejeu de données, qui consiste à ré-entraîner le modèle avec des images d'anciennes classes stockées dans une mémoire externe. Cependant, cette méthode est peu adaptée lorsque la capacité de stockage et la capacité de calcul pour l'entraînement du modèle sont limitées. Nous proposons une nouvelle méthode qui repose sur un schéma d'entraînement basé sur un classificateur un-contre-tous pour répondre à cette limitation. Notre méthode appelée ILOVA (Incremental Learning of One-Vs-All classifiers) offre un meilleur compromis entre précision, oubli, temps de calcul et empreinte mémoire, que les méthodes de l'état de l'art et se montre particulièrement efficace avec des mémoires très réduites, jusqu'à une seule image par classe.En détection d'objets, de nombreux scénarios de test sont construits à partir de séquences vidéo réelles dans lesquelles des objets peuvent réapparaître à plusieurs reprises à différents moments dans le flux de données. Cependant, ce phénomène de revisite que nous appelons rejeu naturel, est peu documenté et son impact sur les performances et l'oubli reste mal compris. Nous proposons alors une métrique, appelée NRS (Natural Replay Score), permettant de quantifier le degré de rejeu naturel d'un scénario, et montrons qu'il est impossible d'évaluer correctement l'oubli d'un modèle en sa présence. La suite de notre étude porte sur l'analyse de l'oubli dans l'architecture Faster R-CNN lorsqu'elle est utilisée en détection d'objets en ligne. D'une part, nos résultats montrent que des rappels réguliers réduisent l'oubli. D'autre part, nous proposons un nouveau protocole, appelé Module Probing, qui permet de mesurer localement l'oubli dans l'architecture. Nous montrons que l’oubli est concentré dans la couche de classification du Faster R-CNN. Enfin, ces analyses nous mènent à proposer la méthode Configurable Recall, basée sur le rejeu de données. Notre méthode permet d'optimiser la fréquence et la durée des rappels et utilise une fonction de perte modifiée qui limite l'oubli dans la couche de classification. En combinant ces deux éléments, nous parvenons à réduire significativement l'oubli pour l’architecture Faster R-CNN
In this thesis, we focus on the problem of online continual learning in artificial neural networks, which involves learning continuously from a data stream. The main challenge is that integrating new information from the stream tends to overwrite previously acquired knowledge, a phenomenon known as catastrophic forgetting.In the field of online continual learning, our research focus on two important applications in computer vision: image classification and object detection. In these cases, the data stream consists of a sequence of images.In image classification, the neural network must progressively learn to classify images from new classes without forgetting the previous ones. The most common method to address this problem is experience replay, which involves retraining the model with images from previously seen classes stored in external memory. However, this method is less suitable when both storage capacity and computational resources are limited. We propose a new method based on a one-vs-all classifier training scheme to overcome this limitation. Our method, called ILOVA (Incremental Learning of One-Vs-All classifiers), offers a better trade-off between accuracy, forgetting, computational time, and memory footprint compared to state-of-the-art methods and proves particularly effective with very limited memory, down to a single image per class.In object detection, many test scenarios are constructed from real video sequences in which objects can reappear multiple times at different moments in the data stream. However, this phenomenon of reappearance, which we call natural replay, is poorly documented, and its impact on performance and forgetting remains poorly understood. We propose a new metric, called NRS (Natural Replay Score), which quantifies the degree of natural replay in a scenario, and show that it is impossible to properly evaluate model forgetting in its presence. The next part of our study focuses on analyzing forgetting in the Faster R-CNN architecture when used for online object detection. On the one hand, our results show that periodic recalls reduce forgetting. On the other hand, we propose a new protocol, called Module Probing, which allows us to measure forgetting locally within the architecture. We show that forgetting is concentrated in the classification layer of Faster R-CNN. Finally, these analyses lead us to propose a method called Configurable Recall, based on experience replay. Our method optimizes the frequency and duration of the recalls and uses a modified loss function to limit forgetting in the classification layer. By combining these two elements, we significantly reduce forgetting in the Faster R-CNN architecture

Les institucions d'ensenyament superior s'han hagut d'adaptar cada vegada més a la digitalització de l'educació, així com a una àmplia demografia d'estudiants que es tornen a incorporar a la formació acadèmica en diferents etapes de la seva carrera. Aquests estudiants, que aprenen al llarg de la vida, sovint han de combinar els estudis amb responsabilitats professionals paral·leles. El propòsit d'aquest estudi interpretatiu és comprendre les experiències dels estudiants en diferents contextos -des d'una perspectiva d'ecologies d'aprenentatge (learning ecologies, LE)- en el context de l'ensenyament superior en línia. L'estudi de cas múltiple de mètodes mixtos principalment qualitatius es va desenvolupar en tres programes de postgrau totalment en línia. Els resultats obtinguts mitjançant l'ús d'un marc analític de LE han demostrat l'efectivitat del constructe a l'hora d'analitzar la complexitat de l'aprenentatge en diferents contextos. Els resultats destaquen la centralitat de l'activitat de l'alumne com un component clau que impulsa les LE d'un individu en combinació amb el suport dels companys i els recursos d'aprenentatge digital en sistemes oberts, dinàmics i fluids que abasten múltiples contextos.
Las instituciones de educación superior han tenido que adaptarse cada vez más a la digitalización de la educación, así como a una amplia demografía de estudiantes que vuelven a incorporarse a la educación formal en diferentes etapas de su carrera. Estos estudiantes, que aprenden a lo largo de su vida, a menudo deben combinar los estudios con responsabilidades profesionales paralelas. El propósito de este estudio interpretativo es comprender las experiencias de los estudiantes en distintos contextos -desde una perspectiva de ecologías de aprendizaje (learning ecologies, LE)- en el contexto de la educación superior en línea. El estudio de caso múltiple de métodos mixtos principalmente cualitativos se desarrolló en tres de programas de posgrado totalmente en línea. Los resultados obtenidos mediante el uso de un marco analítico de LE han demostrado la efectividad del constructo a la hora de analizar la complejidad del aprendizaje en distintos contextos. Los resultados destacan la centralidad de la actividad del alumno como un componente clave que impulsa las LE de un individuo en combinación con el apoyo de los compañeros y los recursos de aprendizaje digital dentro de sistemas abiertos, dinámicos y fluidos que abarcan múltiples contextos.
Higher Education institutions have increasingly had to adapt to both the digitalization of education and a broad global demographic of students re-entering formal education at different stages of their careers as lifelong learners, often in combination with parallel professional responsibilities. The purpose of this interpretive study is to understand student experiences and conceptions of learning across contexts - from a learning ecologies perspective - in the context of online HE. The primarily qualitative mixed methods multiple case study was developed across three sites of fully online graduate level programs (master's or 1st year doctoral students) at the UOC and at UIUC. The results obtained using a Learning Ecologies (LE) analytical framework have demonstrated the effectiveness of the construct for analyzing the complexity of learning across multiple contexts. The findings highlight the centrality of learner activity as a key component which drives an individual's LE in combination with peer support and digital learning resources within open, dynamic and fluid systems spanning multiple contexts.

Orientador: Miriam Godoy Penteado
Banca: João Pedro Mendes da Ponte
Banca: Marcelo de Carvalho Borba
Banca: Maria Elizabeth Bianconcini Trindade Morato Pinto de Almeida
Banca: Vani Moreira Kenski
Resumo: A presente pesquisa analisa a natureza da aprendizagem matemática em um curso online de formação continuada de professores, denominado Geometria com Geometricks. Nele, alunos-professores de uma mesma rede de escolas, situadas em diferentes localidades do país, desenvolveram atividades de Geometria utilizando-se do software Geometricks, e se encontravam para discuti-las. Esses encontros aconteceram a distância, em tempo real, por chat ou videoconferência. Nessa proposta pedagógica, a telepresença condicionou a comunicação e oportunizou o estar-junto-virtual-com-mídias. De modo singular, os recursos da videoconferência permitiram que construções geométricas fossem compartilhadas visualmente e realizadas por todos os envolvidos, fomentando a interação e a participação ativa, constituindo, por meio do diálogo, uma comunidade virtual de aprendizagem. Os resultados levam a inferir que, nesse contexto, a aprendizagem matemática teve natureza colaborativa, na virtualidade das discussões, tecidas a partir das contribuições de todos os participantes; coletiva, na medida em que a produção matemática era condicionada pelo coletivo pensante de seres-humanos-com-mídias; e argumentativa, uma vez que conjecturas e justificativas matemáticas se desenvolveram intensamente do decorrer do processo, contando para isso com as tecnologias presentes na interação ocorrida de forma constante e colaborativa.
Abstract: This study was conducted to analyze the nature of mathematical learning in an online continuing education course for teachers entitled Geometry with Geometricks. Teachers employed in a nation-wide network of privately-supported schools developed geometry activities using the software Geometricks and discussed them in virtual meetings, in real time, via chat or video-conference. In this pedagogical proposal, tele-presence conditioned the communication and provided the opportunity for virtual-togetherness-with-media. In a unique way, the resources of the videoconference made it possible for everyone to participate in and visually share geometrical constructions, encouraging interaction and active participation and constituting a virtual learning community through dialogue. The results indicate that, in this context, mathematical learning nature was characterized by: collaboration, in the virtual discussions that were woven from the contributions of all the participants; collectivity, to the degree to which mathematical production was conditioned by the humans-with-media thinking collective; and argumentation, as the development of mathematical conjectures and justifications was intense throughout the process, aided by the technologies that were present in the constant, collaborative interaction.
Doutor

Made available in DSpace on 2014-06-11T19:31:43Z (GMT). No. of bitstreams: 0 Previous issue date: 2007-03-30Bitstream added on 2014-06-13T18:42:46Z : No. of bitstreams: 1 zulatto_rba_dr_rcla.pdf: 1418316 bytes, checksum: 8cfc1b5fe211e92399513a0cac71bc8b (MD5)
Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)
A presente pesquisa analisa a natureza da aprendizagem matemática em um curso online de formação continuada de professores, denominado Geometria com Geometricks. Nele, alunos-professores de uma mesma rede de escolas, situadas em diferentes localidades do país, desenvolveram atividades de Geometria utilizando-se do software Geometricks, e se encontravam para discuti-las. Esses encontros aconteceram a distância, em tempo real, por chat ou videoconferência. Nessa proposta pedagógica, a telepresença condicionou a comunicação e oportunizou o estar-junto-virtual-com-mídias. De modo singular, os recursos da videoconferência permitiram que construções geométricas fossem compartilhadas visualmente e realizadas por todos os envolvidos, fomentando a interação e a participação ativa, constituindo, por meio do diálogo, uma comunidade virtual de aprendizagem. Os resultados levam a inferir que, nesse contexto, a aprendizagem matemática teve natureza colaborativa, na virtualidade das discussões, tecidas a partir das contribuições de todos os participantes; coletiva, na medida em que a produção matemática era condicionada pelo coletivo pensante de seres-humanos-com-mídias; e argumentativa, uma vez que conjecturas e justificativas matemáticas se desenvolveram intensamente do decorrer do processo, contando para isso com as tecnologias presentes na interação ocorrida de forma constante e colaborativa.
This study was conducted to analyze the nature of mathematical learning in an online continuing education course for teachers entitled Geometry with Geometricks. Teachers employed in a nation-wide network of privately-supported schools developed geometry activities using the software Geometricks and discussed them in virtual meetings, in real time, via chat or video-conference. In this pedagogical proposal, tele-presence conditioned the communication and provided the opportunity for virtual-togetherness-with-media. In a unique way, the resources of the videoconference made it possible for everyone to participate in and visually share geometrical constructions, encouraging interaction and active participation and constituting a virtual learning community through dialogue. The results indicate that, in this context, mathematical learning nature was characterized by: collaboration, in the virtual discussions that were woven from the contributions of all the participants; collectivity, to the degree to which mathematical production was conditioned by the humans-with-media thinking collective; and argumentation, as the development of mathematical conjectures and justifications was intense throughout the process, aided by the technologies that were present in the constant, collaborative interaction.

Made available in DSpace on 2015-07-22T20:49:49Z (GMT). No. of bitstreams: 0 Previous issue date: 2008-11-28
Leite, Maria Teresa Meirelles. Cursos em educação médica continuada on-line na América Latina: um olhar pedagógico. [Online Continuing Medical Education Courses in Latina America: a pedagogical approach]. Orientador: Daniel Sigulem. São Paulo: s.n., 2008. [146]. Dissertação(Mestrado em Ciências)-Universidade Federal de São Paulo. Escola Paulista de Medicina. Resumo: Introdução: A educação médica continuada on-line oferecida por universidades parece priorizar os conteúdos transmitidos, a qualidade estética e a sofisticação tecnológica, dedicando pouca atenção aos aspectos pedagógicos. Os objetivos desta pesquisa são: sistematizar aspectos teóricos de educação, de educação a distância e de educação médica continuada (EMC) e identificar as abordagens pedagógicas prevalentes nos cursos em EMC on-line no Brasil e América Latina, por meio de artigos publicados. Metodologia: Após a elaboração do quadro teórico e do estabelecimento do escopo da pesquisa foram conduzidas buscas bibliográficas em bases de dados eletrônicas internacionais em saúde, medicina e humanidades. A seguir, foram realizadas pesquisas complementares, em websites de universidades da América do Sul. Resultados: Apenas quatro relatos obedeciam plenamente aos critérios estabelecidos para a investigação, sendo que três (3) parecem obedecer à abordagem cognitivista e um (1) à abordagem tradicional. Após a busca complementar, constatou-se que de um total de duzentas e vinte e seis (226) escolas médicas presentes em treze (13) países da América do Sul, estão disponíveis para consulta na Internet treze (13) experiências de EMC on-line. Entre elas, houve referência a concepções de educação em apenas oito (8). Discussão: Foram discriminados os limites impostos a esta pesquisa, observaram-se as características de uma busca bibliográfica envolvendo a área de humanas, estabeleceu-se que os resultados não constituem um referencial para avaliar os cursos quanto à concepção de educação, considerou-se a legitimidade dos resultados como amostragem significativa para ilustrar o volume de artigos disponíveis sobre o tema. Foram analisados os artigos incluídos, assim como diversos achados relevantes. Foram discutidos outros aspectos relacionados ao quadro teórico inicial, como perfil do docente e pós-modernidade. Conclusões: 1. Praticamente inexiste, na comunidade acadêmica latino-americana que desenvolve estudos pedagógicos nesta área, um envolvimento com a publicação científica nas bases de dados pesquisadas. 2. A falta de informações sobre aspectos pedagógicos pode revelar a ausência de um especialista em educação no planejamento dos cursos e pode apontar para a necessidade dos cursos de Pedagogia brasileiros investirem na formação do profissional capaz de atender essa demanda. 3. Há grande tendência das instituições privadas oferecerem educação continuada a distância, o que ficou evidente a partir da pesquisa complementar nos sites das universidades. 4. A partir dos resultados, foi possível constatar também que vários autores concordam que, com as recentes mudanças sociais, muitas transformações são esperadas da Educação. 5. O quadro teórico desenvolvido pode auxiliar a estabelecer fundamentos para reconhecer onde o conhecimento pedagógico se insere no contexto da EaD. 6. Além da pesquisa sobre o número de cursos oferecidos na América Latina e suas características, pode ser relevante utilizar o mesmo referencial teórico para estabelecer critérios de análise qualitativa dos cursos oferecidos por uma instituição específica, indicando possibilidade de novas pesquisas..
Introduction: Continuing medical education online offered by universities in Brazil seems to prioritize aesthetic quality and technological sophistication, paying little attention to pedagogical aspects. The objectives of this research are: to systematize theoretical aspects to the concepts of education, distance learning and continuing medical education (CME) and to identify the pedagogical approaches adopted in CME online courses in Brazil and Latin America by means of published articles. Method: After the construction of the theoretical basis and the establishment of the research limits, bibliographical queries in international electronic databases in health, medicine and humanities were carried through. Following, complementary research was performed in South American universities websites. Results: Only four articles obeyed the established criteria, from which three (3) seem to hold a cognitivist pedagogical approach and one (1) held a traditional approach of education. After the complementary search, it was found that, from a total of two hundred and twenty and six (226) medical schools in thirteen (13) South American countries, there were thirteen (13) online CME courses available for consultation in the Internet. Among them, only eight (8) referred to the education conceptions. Discussion: The limits for the query were discriminated, the characteristics of a bibliographical query involving the humanities was observed, it was established that the present results cannot constitute a referential tool to evaluate the courses, the legitimacy of the results as a significant sampling to illustrate the volume of available articles on the subject was discussed, the four articles were deeply analyzed, as well as diverse other findings that were not included. Besides, other aspects were discussed, as teacher profile and post modernity. Conclusions: 1. There is practically no involvement with scientific publication in the searched databases by the Latin American researchers in educational approaches for online continuing medical education. 2. The lack of information on pedagogical aspects may reveal the absence of an education specialist in the courses design and may suggest for the necessity for Brazilian Pedagogy courses to form a professional who is capable of answering this demand. 3. The trend of private institutions to offer online CME was evidenced from the complementary research. 4. From the results, it was possible to see that some authors agree that with the recent social changes, many transformations are expected from Education. 5. The developed theoretical basis may be able to contribute to establish connections to understand how the pedagogical knowledge inserts in the context of distance learning. 6. Beyond the research on the number of courses offered in Latin America and its characteristics, it is desirable to use the same theoretical referential to establish criteria for qualitative analysis for the courses offered by a specific institution, indicating a possibility of continuing this research.
TEDE
BV UNIFESP: Teses e dissertações

Este estudo teve como objetivo investigar a viabilidade da educação online como estratégia de formação continuada de professores de música em exercício na educação básica. O método de pesquisa utilizado foi a pesquisa-ação, que consistiu de um curso online sobre tecnologia na educação musical destinado a professores de música de escolas públicas situadas no Estado do Pará. Oito professores aceitaram participar da pesquisa. Foram utilizadas três técnicas de pesquisa: questionário auto-administrado, observação participante e entrevista semiestruturada. Os resultados da presente pesquisa indicam que as bases para criação de programas de formação continuada de professores de música em exercício na educação básica, por meio da educação online, estão na concepção metodológica que a orienta, na gestão eficiente do ambiente, na adequação do material didático disponibilizado ao conteúdo do curso e no apoio institucional.
This study aimed to investigate the viability of online education as a strategy for inservice music teacher education. The research method used was the actionresearch, which consisted of an online course on music technology for in-service music teachers of public schools situated in the State of Pará, Brazil. Eight teachers agreed to participate in this research. Three research techniques have been used: self-administered questionnaires, observation and semi-structured interviews. The results point out that the basis for the creation of in-service music teacher education programs, using online education, resides on the conception that orient the program, on efficient administration of the environment, on adequation of the teaching material available to the content of the course and on the institutional support.

Aquesta recerca intenta trobar una definició d'abandonament (dropout) empírica i adaptada als estudiants de la UOC i, per extensió, potencialment adaptable a altres institucions d'aprenentatge a distància en línia. Hem fet una anàlisi de la quantitat de semestres consecutius de descans (N) que fan que es consideri que un estudiant d'un programa determinat ha abandonat els estudis (amb una probabilitat inferior al 5% de retorn o error màxim). Aquesta N és diferent per a cada programa, com també ho és el percentatge global d'abandonament. Per a tots els programes hi ha una alta probabilitat d'abandonament després d'una pausa, especialment quan aquesta interrupció és durant el segon o tercer semestre (early dropout). La recerca continua posant el focus d'anàlisi en l'early dropout, amb la definició de dos models per al conjunt de nous estudiants de grau que descansen durant el segon semestre: el model d'intenció de continuar i el de rematrícula efectiva (o prolongació de descans), ambdós durant el tercer semestre.
Esta investigación trata de encontrar una definición de abandono (dropout) empírica y adaptada a los estudiantes de la UOC y, por extensión, potencialmente adaptable a otras instituciones de aprendizaje a distancia en línea. Hemos realizado un análisis de la cantidad de semestres consecutivos de descanso (N) que hacen que se considere que un estudiante de un programa determinado ha abandonado los estudios (con una probabilidad inferior al 5 % de retorno o error máximo). Esta N es diferente para cada programa, así como el porcentaje global de abandono. Para todos los programas, hay una alta probabilidad de abandono tras una pausa, especialmente cuando esta interrupción es en el segundo o tercer semestre (early dropout). La investigación continúa poniendo el foco de análisis en el early dropout, con la definición de dos modelos para el conjunto de nuevos estudiantes de grado que descansan en el segundo semestre: el modelo de intención de continuar y el de rematrícula efectiva (o prolongación de descanso), ambos en el tercer semestre.
The aim of this research is to establish an empirical definition of dropout that is tailored to UOC students and can be adapted later on to other online distance learning institutions. To do so, we analyse the number of consecutive rest periods (N) needed for a student on a particular programme to be considered to have dropped out (with a probability of less than 5% return or maximum error). This N proves to be different for each programme, as does the overall dropout rate. All programmes show a high probability of students' leaving after a break, especially when this interruption in their studies takes place in the second or third semester (early dropout). The research then focuses on early dropout, defining two models for the total number of new bachelor's degree students taking a break in the second semester: the continuance intention model and the effective re-enrolment model (or extended rest period), both for the third semester.

Made available in DSpace on 2016-12-08T16:35:13Z (GMT). No. of bitstreams: 1 Soraya Tonelli.pdf: 1483282 bytes, checksum: d49dcee4bf8aebbce3a37709d4d30f54 (MD5) Previous issue date: 2011-03-15
Coordenação de Aperfeiçoamento de Pessoal de Nível Superior
This investigative work was subject to avoidance in distance learning courses online. In this sense was conducted a case study of the Faça e Aconteça Program held by Sebrae in Santa Catarina in the corresponding period of the second half of 2009 and first half of 2010. Were identified the factors that contribute to dropout in distance learning courses online as part of continuing non-formal adult education; the profile of persistent and dropouts students in terms of demographics, habits, preferences and study conditions; and the relationship between the students' profile and factors of permanence or abandonment of the program. The study results were obtained using two main procedures of data collection: a quantitative survey applied to a random sample of two hundred participants of the Faça e Aconteça Program and a qualitative deepening through focused individual interviews applied to a non-probability sample of ten participants of the Program, both procedures involving those who completed the courses and those who have not finished. Survey participants were selected from the base of subscribers who effectively started the courses in the program studied in the period already mentioned above, compounding a research universe consists of 1,347 people distributed in three categories: (1) those who completed the six courses of the program (634 persons) (2) those who completed at least one course of the program (379 people), and (3) those who have not completed any of the six courses (334 people). From the base sample composed of 1,347 participants was selected for the quantitative study a simple random sample of 200 people, resulting in a margin of error of + / - 7.6%, considering a confidence level of 95.5% . The survey instrument used for quantitative data collection was based on the instrument originally developed by researcher Amanda Moura Walter in 2006 and refined and validated by the researcher Onília de Almeida in 2007. Descriptive statistical analysis were made with data from the quantitative part, and content analysis was performed with data from the qualitative part. From the analysis of results was possible to infer that there are differences in the profile of students that influence the persistence or dropout, and that the main factors contributing to the evasion can be classified as: causes related to the program itself, related to situational factors, lack of planning for the study, and/or personal difficulty in adapting to distance education
Este trabalho teve como temática investigativa a evasão em cursos a distância online. Neste sentido se realizou um estudo de caso do Programa Faça e Aconteça promovido pelo Sebrae em Santa Catarina no período correspondente ao segundo semestre de 2009 e primeiro semestre de 2010. Foram identificados os fatores que contribuem para a evasão em cursos a distância online no âmbito da educação continuada não formal de adultos; o perfil dos estudantes persistentes e desistentes em termos de aspectos demográficos, hábitos, preferências e condições de estudo; e a relação existente entre o perfil dos estudantes e fatores de permanência ou desistência do Programa. Os resultados do estudo foram obtidos por meio de dois procedimentos principais de coleta de dados: uma pesquisa quantitativa aplicada a uma amostra probabilística de duzentos participantes do Programa Faça e Aconteça, e um aprofundamento qualitativo por meio de entrevista individual focada aplicada a uma amostra não probabilística de dez participantes do Programa, ambos os procedimentos envolvendo tanto concluintes quanto não concluintes. Os participantes da pesquisa foram selecionados a partir da base de inscritos que efetivamente iniciaram os cursos do Programa estudado no período já assinalado mais acima, compondo um universo de pesquisa formado por 1.347 pessoas distribuídas em três categorias: (1) aquelas que concluíram os seis cursos do Programa (634 pessoas); (2) aquelas que concluíram pelo menos um curso do Programa (379 pessoas); e (3) aquelas que não concluíram nenhum dos seis cursos (334 pessoas). A partir da base amostral composta pelos 1.347 participantes foi selecionada para a pesquisa quantitativa uma amostra probabilística simples de 200 pessoas, o que resulta em uma margem de erro de +/- 7,6%, considerando um nível de confiança de 95,5%. O instrumento de pesquisa utilizado para levantamento dos dados quantitativos foi baseado no instrumento originalmente desenvolvido pela pesquisadora Amanda Moura Walter no ano de 2006 e validado e refinado pela pesquisadora Onília de Almeida no ano de 2007. Com os dados da parte quantitativa foram feitas análises estatísticas descritivas, e com os dados da parte qualitativa foi realizada análise de conteúdo. A partir da análise dos resultados foi possível inferir que existem diferenças no perfil dos estudantes que influenciam na persistência ou na evasão, e que os principais fatores que contribuem para a evasão podem ser classificados em: causas relativas ao próprio programa, causas que dizem respeito a fatores situacionais, causas afetas à falta de planejamento para o estudo, e/ou causas relacionadas à dificuldade pessoal de adaptação à educação a distância

L'intelligence artificielle (IA) est une technique puissante pour créer de la valeur et améliorer la prise de décisions. Les volumes de données générés devraient connaître une croissance exponentielle, avec des projections dépassant les 2 000 zettaoctets d'ici 2035, soulignant l'importance de l'apprentissage incrémental et des modèles scalables pour le ”Big ` Data”. L'apprentissage et l'entraînement continus des modèles d'IA, connus sous le nom de Streaming ` Machine Learning (SML), sont des approches efficaces permettant de construire et de mettre à jour des modèles d'intelligence artificielle au fur et à mesure de l'arrivée de nouvelles données, sans avoir besoin de remplacer périodiquement le modèle complet. La détection des changements de comportement en temps réel est une solution pertinente pour détecter des évènements critiques et à diverses applications telles que la détection de fraudes, de cyberattaques ou de déviations dans les systèmes d'information. Les travaux de cette recherche visent donc à résoudre les défis algorithmiques et d'infrastructure liés aux modèles d'IA à apprentissage continu et à l'industrialisation des modèles dynamiques sur des flux de données issus de sources diverses. Les contributions de cette thèse incluent StreamFlow pour résumer des flux de données massives, Stream2Graph pour construire et mettre à jour dynamiquement des graphes de connaissances, et StreamChange pour détecter et expliquer des déviations ou changements de comportement en temps réel. Les expériences sur des données réelles démontrent que notre modèle StreamChange détecte mieux les changements graduels et brusques que les modèles de l'état de l'art. Enfin, la thèse introduit une architecture, StreamM- ` LOps, pour l'apprentissage en ligne et l'industrialisation des modèles, évaluée sur des données dynamiques contenant des millions de variables pour la détection en ligne d'évènements malveillants. Nous concluons par un retour d'expérience sur la mise en production et la maintenance des modèles évolutifs (SML) pour des applications de temps réel en banque
Artificial Intelligence (AI) is a powerful tool to extract valuable insights for decision-making. However, learning from heterogeneous and unstructured streaming data presents a multitude of challenges that this research aims to tackle. The creation of big data is projected to experience exponential growth, with expectations to surpass 2,000 zettabytes by the year 2035. Such Big Data highlights the importance of efficient, incremental, and adaptive models. Online Learning, known as Streaming Machine Learning (SML), is a dynamic technique for building and updating learning models as new data arrive, without the need for periodic complete model replacement. It is the most efficient technique for big data stream learning. The change detection task is a proactive way to detect and prevent critical events such as cyber-attacks, fraud detection, or IT incidents in an online fashion. The research conducted during this thesis aims to develop adaptive and scalable online machine-learning solutions to learn from heterogeneous streaming data that can be operationalized with large-scale infrastructures, particularly in the banking sector. This Ph.D. thesis delves into algorithmic and infrastructure challenges related to continuous training and serving online machine learning over high-velocity streaming data from diverse sources, specifically focusing on large-scale IT infrastructures (AIOps). Thesis contributions include techniques like StreamFlow for summarizing information from big data streams, Stream2Graph for dynamically building and updating knowledge graphs for batch and online learning tasks, and StreamChange, an efficient and explainable online change detection model. Evaluation results on real-world open data and industrial data demonstrate performance improvements in learned models. StreamChange surpasses state-ofthe-art techniques in detecting gradual and abrupt changes. Additionally, the thesis introduces a conceptual framework, StreamMLOps, for scaling and serving online machine learning in real-time without pausing the inference pipeline. This framework showcases the effectiveness of the proposed MLOps pipeline on a feature-evolving dataset with millions of dimensions for malicious event detection tasks. Finally, we share lessons learned regarding Streaming Machine Learning systems, AI at scale, and online model management in large-scale banking, with a focus on streaming data and real-time applications

Made available in DSpace on 2016-06-02T19:35:30Z (GMT). No. of bitstreams: 1 1640.pdf: 4365691 bytes, checksum: 451de24779c6844de902661f5dd98e13 (MD5) Previous issue date: 2007-12-12
This doctorate thesis identifies and analyzes the contributions of two online distance minicourses to the knowledge base (specific knowledge of geometry, pedagogical knowledge of geometry and pedagogical reasoning processes) of thirteen elementary school teachers who teach geometry from the 1st to the 4th grades. They were participating in a teachers training program called Programa de Apoio aos Educadores: espaço de desenvolvimento profissional, and were members of Portal dos Professores UFSCar (www.portaldosprofessores.ufscar.br), a website which provides teachers with professional support. It is a qualitative intervention research which employs a collaborative and constructive methodology for training teachers. The mini-courses, developed for that purpose, were given totally at distance through the virtual learning environment WebCT. The theoretical basis encompasses teachers training, knowledge base for teaching and distance learning via the Internet as a resource for training teachers. The data, which were collected through digital tools (forum, e-mail, chat room, tasks, message board etc.) and non-digital tools (application forms and questionnaires), were analyzed to show the impact the mini-courses had on the teachers` knowledge base. The results showed their knowledge was significantly enhanced and they could think over the teaching of geometry and their own teaching. The results also showed the strong and weak points of training teachers at distance using the virtual learning environment WebCT and the methodology adopted. To conclude, the study offers suggestions for public policies on distance training for teachers via the Internet and for courses in virtual learning environments similar to WebCT
Este estudo identifica e analisa as contribuições de dois Minicursos à distância, via Internet, à base de conhecimento (Conhecimento do Conteúdo Específico de Geometria, Conhecimento Pedagógico do Conteúdo de Geometria e Processos de Raciocínio Pedagógico) de treze professoras que ensinam Geometria da 1ª à 4ª série do Ensino Fundamental, associadas ao "Programa de Apoio aos Educadores: Espaço de Desenvolvimento Profissional" e ao site Portal dos Professores da UFSCar (www.portaldosprofessores.ufscar.br). Nesta pesquisaintervenção, de natureza qualitativa, foi adotada a metodologia construtivo-colaborativa para a formação continuada de professores. Os Minicursos, desenvolvidos para esse fim, foram implementados totalmente à distância, por meio do ambiente virtual de aprendizagem WebCT. O referencial teórico abarca a aprendizagem profissional da docência, a base de conhecimento para o ensino e a Educação a Distância via Internet como recurso para a formação de professores. Os dados, coletados por meio de ferramentas digitais (fórum, correio eletrônico, bate-papo, tarefas/atividades, quadro de comunicações etc.) e não digitais (fichas de inscrição e questionários), foram analisados com vistas à verificação do impacto dos Minicursos na base de conhecimento das professoras-alunas. Os resultados demonstraram que elas ampliaram significativamente sua base de conhecimento para o ensino de Geometria e tiveram a oportunidade de refletir, de forma crítica e fundamentada, sobre o ensino de Geometria e sobre a própria prática pedagógica, modificando-a de forma sustentada. Os resultados indicaram, também, a potencialidade, as dificuldades e os limites da formação de professores a distância por meio do ambiente WebCT e da metodologia utilizada. Na conclusão, são apresentadas sugestões para políticas públicas de formação inicial e continuada de professores, no âmbito da Educação a Distância via Internet, e para o desenvolvimento de cursos em ambientes virtuais de aprendizagem com características similares às do WebCT

Made available in DSpace on 2017-07-21T20:31:53Z (GMT). No. of bitstreams: 1 Carlos Roberto Ferreira.pdf: 807896 bytes, checksum: 223cca163278b33a9f78114a7d011643 (MD5) Previous issue date: 2010-11-25
Fundação Araucária de Apoio ao Desenvolvimento Científico e Tecnológico do Paraná
This paper aimed to clarify the question about what becomes evident in the Mathematical Modeling course, offered in the Distance Education online method for the continuing education of mathematics teachers. The central aim of this research was to understand how mathematical modeling developed in an online course on distance education modality can contribute to overcome teacher‟s difficulties in understanding the methodology and its use in the classroom. Also, we aimed to ascertain, among other things, which contributions to modality online distance learning provides for the teacher‟s continuing education and to in the difficulties superation of adoption and understanding of the mathematical modeling concepts. The theoretical framework considers teaching and learning of mathematics perspectives in the context of the “Leis de Diretrizes e Bases” of National Education. It also holds in most current views on mathematics education and mathematics teaching that are present in the work of Lorenzato and Fiorentini (2001), Rius (1989) and Kilpatrick (1992). For the concepts of mathematical modeling attempted to talk to different authors, however, we assume the concept proposed by Burak (2004). What refers to the teacher‟s continuous formation, we adopted a "reflective practice" based on the work of Schön (2000), Silva and Araujo (2004), Freire (2001), Perrenoud (2002), Imbernon (2001), Alarcão (2003). About the Distance Learning (DL) it drives us in the work of Borba (2007), Mugnolo (2009) and Morin (2002) and the Online Distance Education in the concept named "virtual togetherness” according to Valente (2003). Other distinguished authors and theoretical elements were added to the discussions as incursions were necessary to understand the constructed categories. The question and the objectives of the investigation leaded it to qualitative research. Twelve teachers of public schools in Paraná joined in the investigation, involving five Regional Centers of Education. The steps followed in the investigation included: 1) exploratory research with teachers and construction of theoretical frameworks, 2) Mathematical Modeling course development mediated by the online distance education, that allowed the collection of data from the Moodle Platform tools, 3) analysis and data interpretation. The data treatment follows the method of triangulation and inductive analysis, involving the researcher, empirical data and theoretical referential, by the light of the Bogdan and Biklen‟s principles (1994). The results show significant aspects regarding the development of Mathematical Modelling and its adoption in Distance Education in Basic Education, although, it highlights the need for adjustments in the dynamics of the course and aspects related to the modeling concept from a perspective of the humanities and social sciences.
Esta investigação buscou explicitar a questão sobre o que se evidencia num curso de Modelagem Matemática, oferecido na modalidade de Educação a Distância online, para a formação continuada do professor de matemática. O objetivo central da investigação consistiu em compreender como a Modelagem Matemática desenvolvida num curso na modalidade EaD online pode contribuir para a superação das dificuldades do professor no entendimento da metodologia e na sua utilização em sala de aula. Também objetivamos verificar, entre outros pontos, quais contribuições a modalidade EaD online proporciona para a formação continuada dos professores e na superação das dificuldades de adoção e entendimento das concepções da Modelagem Matemática. O referencial teórico contempla as perspectivas para o ensino e aprendizagem da Matemática no contexto da Lei de Diretrizes e Bases da Educação Nacional. Sustenta-se ainda nas visões mais atuais sobre Educação Matemática e o ensino de matemática que estão presentes nos trabalhos de Lorenzato e Fiorentini (2001), Rius (1989) e Kilpatrick (1992). Quanto às concepções de Modelagem Matemática buscou-se dialogar com diversos autores, no entanto, assumimos a concepção proposta por Burak (2004). No tocante à formação continuada de professores adotamos a "pratica reflexiva" como base nos trabalhos de Schön (2000), Silva e Araújo (2004), Freire (2001), Perrenoud (2002), Imbernón (2001), Alarção (2003). No que se refere à Educação a Distância (EaD) nos pautamos nos trabalhos de Borba (2007), Mugnol (2009) e Morin (2002) e na EaD Online no conceito denominado “estar junto virtual de acordo com Valente (2003). Outros elementos teóricos e distintos autores foram agregados às discussões conforme se mostraram necessárias incursões para a compreensão das categorias construídas. A questão e os objetivos da investigação a direcionam para uma pesquisa de natureza qualitativa. Participaram da investigação doze professores da rede estadual de ensino do Paraná, envolvendo cinco Núcleos Regionais de Educação. As etapas percorridas na investigação compreenderam: 1) pesquisa exploratória junto aos professores e construção dos referenciais teóricos; 2) desenvolvimento do curso de Modelagem Matemática mediado pela EaD online, que permitiu a coleta de dados a partir das ferramentas da Plataforma Moodle; 3) análise e interpretação dos dados. O tratamento dos dados segue o método da triangulação e análise indutiva, envolvendo os dados empíricos, pesquisador e referencial teórico, a luz dos princípios de Bogdan e Biklen(1994). Os resultados apresentam aspectos significativos em relação ao desenvolvimento da Modelagem Matemática na EaD e a sua adoção na Educação Básica, no entanto, evidencia a necessidade de ajustes na dinâmica do curso e aspectos relativos à concepção de Modelagem em uma perspectiva das Ciências Humanas e Sociais.

Submitted by Débora Massetto (debora.massetto@gmail.com) on 2018-05-15T15:36:00Z No. of bitstreams: 1 TESE_Debora Cristina Massetto_VERSAO FINAL_15 de maio.pdf: 3223336 bytes, checksum: 8be7bbfd67c940b5a3d1e8a44b8c7d31 (MD5)
Rejected by Ronildo Prado (ri.bco@ufscar.br), reason: Oi Débora, Faltou enviar a Carta comprovante assinada pelo orientador. Solicite o modelo em sua Secretaria de Pós-graduação, preencha e colete a assinatura com o orientador e acesse novamente o sistema para fazer o Upload. Fico no aguardo para finalizarmos o processo. Abraços Ronildo on 2018-05-17T20:27:28Z (GMT)
Submitted by Débora Massetto (debora.massetto@gmail.com) on 2018-06-05T13:50:36Z No. of bitstreams: 2 TESE_Debora Cristina Massetto_VERSAO FINAL_15 de maio.pdf: 3223336 bytes, checksum: 8be7bbfd67c940b5a3d1e8a44b8c7d31 (MD5) carta orientador debora massetto (1).jpg: 1090868 bytes, checksum: 8bf857d83c6982662295b8fbbde08a09 (MD5)
Approved for entry into archive by Ronildo Prado (ri.bco@ufscar.br) on 2018-06-06T20:02:59Z (GMT) No. of bitstreams: 2 TESE_Debora Cristina Massetto_VERSAO FINAL_15 de maio.pdf: 3223336 bytes, checksum: 8be7bbfd67c940b5a3d1e8a44b8c7d31 (MD5) carta orientador debora massetto (1).jpg: 1090868 bytes, checksum: 8bf857d83c6982662295b8fbbde08a09 (MD5)
Approved for entry into archive by Ronildo Prado (ri.bco@ufscar.br) on 2018-06-06T20:05:13Z (GMT) No. of bitstreams: 2 TESE_Debora Cristina Massetto_VERSAO FINAL_15 de maio.pdf: 3223336 bytes, checksum: 8be7bbfd67c940b5a3d1e8a44b8c7d31 (MD5) carta orientador debora massetto (1).jpg: 1090868 bytes, checksum: 8bf857d83c6982662295b8fbbde08a09 (MD5)
Made available in DSpace on 2018-06-06T20:05:24Z (GMT). No. of bitstreams: 2 TESE_Debora Cristina Massetto_VERSAO FINAL_15 de maio.pdf: 3223336 bytes, checksum: 8be7bbfd67c940b5a3d1e8a44b8c7d31 (MD5) carta orientador debora massetto (1).jpg: 1090868 bytes, checksum: 8bf857d83c6982662295b8fbbde08a09 (MD5) Previous issue date: 2018-02-26
Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)
The research has as purpose to investigate, to describe and to analyze the professional learning and emotional experiences of experienced teachers - future mentors expressed during their participation in an online training program. In addition, a professional teacher development program was developed, via the internet, at the UFSCar Portal of Teachers (www.portaldosprofessores.ufscar.br), entitled Programa de Formação Online de Mentores (PFOM). The program focused on the mentors formation, i.e., teachers, pedagogical coordinators, school directors and in-service teaching supervisors accompanying beginning teachers, helping them to minimize the typical difficulties of this stage of the career in their schools or teaching networks. Specifically, it was investigated and understood, together with the experienced teachers, the learning and knowledge produced in the process of building identity as mentors throughout their participation in the Program, and how they helped in the mentoring of beginner teachers; it was analyzed the emotional experiences of the experienced teachers and how they interfered in their work, in the construction of their identity as a mentor and in the construction of the knowledge necessary to act in a new role. The theoretical framework concentrated on the learning literature and professional development of teaching, knowledge base for teaching, human development ecology, mentoring programs, emotional experience and written narratives which have as main authors: Shulman, Garcia, Mizukami, Bronfenbrenner, Freire et al. A constructive-collaborative approach to research and intervention was adopted and the main instruments of data collection were written narratives elaborated by the participants. Data analysis was performed based on the bioecological theory systems model. These analyzes were made through studies of multiple cases derived from the professional learning and the emotional experiences of mentors: a Triad (experienced teacherbeginner teachers) and two dyads (experienced teacher-beginner). The analyzes indicated that the inclusion of experienced teachers in a new context (PFOM) and the systematic accompagnement with the beginners proved to be a potentially relevant strategy for the construction of professional knowledge related to the role of teacher trainers. Additionally, it was evidenced the emotional dimension of mentoring, revealed by the positive and negative emotional experiences associated with mentoring job. While the online training context, the PFOM has proved to be a promising pedagogical tool in the continuing education and a possibility to support the professional development of teachers in different phases of the career. An important contribution of the present study to the educational area was to involve different subjects - researchers, tutors and teachers, in different stages of the career, in developmental processes caused by the insertion in a new context, PFOM, with its own activities, such as the study and theoretical deepening, mentoring, mentoring, demonstrating and considering that, at different levels, they learn and construct different types of knowledge.
A pesquisa teve como propósito investigar, descrever e analisar as aprendizagens profissionais e as experiências emocionais de professoras experientes – futuras mentoras, expressas ao longo de sua participação em um programa de formação online. Além disso, desenvolveu-se um programa de desenvolvimento profissional docente, via internet, no Portal dos Professores da UFSCar (www.portaldosprofessores.ufscar.br), intitulado Programa de Formação Online de Mentores (PFOM). O Programa era voltado para a formação de mentores, isto é, professores, coordenadores pedagógicos, diretores de escola e supervisores de ensino em exercício que acompanhavam professores iniciantes, auxiliando-os a minimizar dificuldades típicas desta etapa da carreira em suas escolas ou redes de ensino. Especificamente, investigou-se e compreendeu-se, junto às professoras experientes, as aprendizagens e os conhecimentos produzidos no processo de construção de identidade como mentoras, ao longo de sua participação no Programa, e como estes auxiliaram na mentoria dos professores iniciantes; analisou-se as experiências emocionais das professoras experientes e como estas interferiram em seu trabalho, na construção de sua identidade como mentora e na construção de conhecimentos necessários para atuarem em um novo papel. O referencial teórico pautou-se na literatura sobre aprendizagem e desenvolvimento profissional da docência, base de conhecimento para o ensino, ecologia do desenvolvimento humano, programas de mentoria, experiência emocional e narrativas escritas, tendo como principais autores: Shulman, Garcia, Mizukami, Bronfenbrenner, Freire e colaboradores. Adotou-se uma abordagem construtivo-colaborativa de pesquisa e intervenção e os principais instrumentos de coleta de dados foram narrativas escritas elaboradas pelas participantes. As análises dos dados foram realizadas processualmente com base no modelo de sistemas da teoria bioecológica. Tais análises foram feitas por meio de estudo de casos múltiplos, decorrentes das aprendizagens profissionais e das experiências emocionais das mentoras, uma Tríade (professora experiente-professoras iniciantes) e duas díades (professora experiente-iniciante). As análises indicaram que a inserção de professores experientes em um novo contexto (PFOM) e o acompanhamento sistemático junto aos iniciantes mostrou-se como uma estratégia potencialmente relevante para a construção de conhecimentos profissionais relacionados ao papel de formadores de professores. Adicionalmente, evidenciaram a dimensão emocional da mentoria, revelada pelas experiências emocionais positivas e negativas ligadas ao trabalho de mentora. Enquanto contexto formativo online, o PFOM mostrou-se uma ferramenta pedagógica promissora na formação continuada e uma possibilidade para apoiar o desenvolvimento profissional de professores em diferentes fases da carreira. Uma contribuição importante do presente estudo para a área educacional foi a de envolver diferentes sujeitos – pesquisadores, tutores e professores, em distintas etapas da carreira, em processos desenvolvimentais causados pela inserção em um novo contexto, o PFOM, com atividades próprias, como o estudo e aprofundamento teórico, tutoria, mentoria, demonstrando e considerando que, em diversos níveis, aprendem e constroem diferentes tipos de conhecimentos.

Submitted by CARLOS ROBERTO DOS SANTOS JÚNIOR null (carlos9_rsj@hotmail.com) on 2017-10-26T17:32:23Z No. of bitstreams: 1 Tese_CarlosRSantosJr_biblioteca_final.pdf: 3193090 bytes, checksum: ce950f71ef02dd650ba662b7b8eea784 (MD5)
Approved for entry into archive by Monique Sasaki (sayumi_sasaki@hotmail.com) on 2017-10-31T18:06:36Z (GMT) No. of bitstreams: 1 santosjunior_cr_dr_ilha.pdf: 3193090 bytes, checksum: ce950f71ef02dd650ba662b7b8eea784 (MD5)
Made available in DSpace on 2017-10-31T18:06:36Z (GMT). No. of bitstreams: 1 santosjunior_cr_dr_ilha.pdf: 3193090 bytes, checksum: ce950f71ef02dd650ba662b7b8eea784 (MD5) Previous issue date: 2017-09-04
A evolução dos recursos de internet levou ao crescente aumento do fluxo de dados, e por consequência, a necessidade de modelos de classificação ou previsão que suportem uma aprendizagem online. A Rede Neural ARTMAP Fuzzy tem sido utilizada nas mais diversas áreas do conhecimento, no entanto, ainda é pouco explorada em aplicações de tempo real que exigem uma aprendizagem contínua. Neste trabalho, é proposto uma Rede Neural ARTMAP Fuzzy com treinamento continuado, capaz de adquirir conhecimento ao longo da classificação ou previsão. Modificações na arquitetura e no algoritmo de aprendizagem possibilitam à rede neural ativar o treinamento sempre que necessário. Para validar o modelo proposto foram realizadas duas aplicações, uma para previsão e outra para classificação, utilizando bases de dados benchmarks e comparado com a ARTMAP Fuzzy original. Os resultados mostraram a capacidade do modelo proposto em adquirir conhecimento ao longo das amostras apresentadas de forma estável e eficiente. Assim, este estudo contribui para a evolução da rede neural ARTMAP Fuzzy e apresenta o treinamento continuado como uma alternativa eficaz para aplicações de tempo real.
The evolution of internet resources has led to an increase in the flow of data, and consequently, the need for classification or forecasting models that support an online learning. The ARTMAP Fuzzy Neural Network has been used in several areas of knowledge, however, it is still little explored in real-time applications that require continuous learning. In this work, an ARTMAP Fuzzy Neural Network with continuous training is proposed, able to acquire knowledge along the classification or prediction. Modifications in the architecture and learning algorithm enable the neural network to activate training whenever necessary. To validate the proposed model two experiments were performed, one for forecasting and another for classification, using benchmark databases and compared with the original ARTMAP Fuzzy Neural Network. The results showed the ability of the proposed model to acquire knowledge along the presented samples in a stable and efficient way. Thus, this study contributes to the evolution of the ARTMAP Fuzzy neural network and presents the continuous training as an effective alternative to real-time applications.

Orientador: Anna Diva Plasencia Lotufo
Resumo: A evolução dos recursos de internet levou ao crescente aumento do fluxo de dados, e por consequência, a necessidade de modelos de classificação ou previsão que suportem uma aprendizagem online. A Rede Neural ARTMAP Fuzzy tem sido utilizada nas mais diversas áreas do conhecimento, no entanto, ainda é pouco explorada em aplicações de tempo real que exigem uma aprendizagem contínua. Neste trabalho, é proposto uma Rede Neural ARTMAP Fuzzy com treinamento continuado, capaz de adquirir conhecimento ao longo da classificação ou previsão. Modificações na arquitetura e no algoritmo de aprendizagem possibilitam à rede neural ativar o treinamento sempre que necessário. Para validar o modelo proposto foram realizadas duas aplicações, uma para previsão e outra para classificação, utilizando bases de dados benchmarks e comparado com a ARTMAP Fuzzy original. Os resultados mostraram a capacidade do modelo proposto em adquirir conhecimento ao longo das amostras apresentadas de forma estável e eficiente. Assim, este estudo contribui para a evolução da rede neural ARTMAP Fuzzy e apresenta o treinamento continuado como uma alternativa eficaz para aplicações de tempo real.
Doutor

La investigació se centra en el procés de canvi del coneixement pedagògic de professors universitaris sobre Avaluació per a l'aprenentatge en el marc d'un programa de formació contínua i en línia. Va ser un estudi quasi experimental i autocontrolat, que va suposar una etapa prèvia i posterior a una intervenció. Entre els principals resultats es destaca un efecte positiu del programa formatiu en el canvi del coneixement pedagògic del professor. Això s'identifica a partir del pas d'un enfocament identificat generalment com convencional cap a un altre enfocament més innovador. Això es va fer evident a partir dels resultats proporcionats pel qüestionari, l'entrevista, una escala de confiança i les activitats formatives. No s'han identificat destacades diferències entre els participants considerant les seves característiques demogràfiques i professionals, amb excepció d'algunes, la qual cosa permet de certa manera precisar que la proposta formativa és viable d'aplicar-la a professors universitaris en general.
La investigación se centra en el proceso de cambio del conocimiento pedagógico de profesores universitarios sobre Evaluación para el aprendizaje en el marco de un programa de formación continua y en línea. Fue un estudio cuasi experimental y autocontrolado, que supuso una etapa previa y posterior a una intervención. Entre los principales resultados se destaca un efecto positivo del programa formativo en el cambio del conocimiento pedagógico del profesor. Ello se identifica a partir del paso de un enfoque identificado generalmente como convencional hacia otro enfoque más innovador. Ello se hizo evidente a partir de los resultados proporcionados por el cuestionario, la entrevista, una escala de confianza y las actividades formativas. No se han identificado destacadas diferencias entre los participantes considerando sus características demográficas y profesionales, con excepción de algunas, lo cual permite de cierta manera precisar que la propuesta formativa es viable de aplicarla a profesores universitarios en general.
The research focuses on the process of changing the educational knowledge of university professors on Assessment for Learning within the framework of a program of continuing education and online. It was a quasi-experimental and self-controlled study, which involved a pre and post-intervention. The main results a positive effect of the training program in changing the pedagogical knowledge of the teacher stands out. This is identified from the passage of a conventional generally identified as a more innovative approach to approach. This was evident from the results provided by the questionnaire, the interview, a level of trust and training activities. We have not identified major differences among participants considering their demographic and professional characteristics, except for some, allowing in some way to specify that the training proposal is feasible to apply to university professors in general.

Made available in DSpace on 2016-06-02T19:35:49Z (GMT). No. of bitstreams: 1 5068.pdf: 5385850 bytes, checksum: 1ffb52e70c7105109032abba43f61689 (MD5) Previous issue date: 2013-03-01
Universidade Federal de Sao Carlos
The present study lies in discussions about teacher education and investigated the learning and teachers professional development by making use of teaching cases and case methods, while training tool and investigative processes driving reflective about the knowledge of teaching. The teaching cases are narratives of episodes usually extracted from real data and daily school experiences that give visibility to the teachers experiences, that when analyze them and build them, articulate and relive experiences, establishing relationships between theory and practice. The research and intervention process took place through the Teachers Portal - UFSCar in progress: "Casos de Ensino e Teorização de Práticas Pedagógicas - professores alfabetizadores [Cases of Teaching and Theorizing Pedagogical Practices - literacy teachers], the prospect of continued training of sixteen teachers in the same stage of elementary school (Early Years - 1st, 2nd and 3rd year), which act as literacy. To accomplishment this study, of qualitative approach, the research aimed to analyze episodes of the professional development process experienced by literacy teachers to make use of teaching cases in the virtual environment, while the possibility formative and investigative. Thus, were worked in cases depict events related to the daily lives of literacy. The achieved results in the research indicate the potential of teaching cases in / for learning and for professional development of teaching. In general, the strategies of analysis and preparation of teaching cases enabled knowledge related to the teaching and learning of language were explained and (re) assessed by literacy. The use of teaching cases and case methods allowed reflections if they were present and provided evidence of the doubts, certainties and contradictions that characterize and guide the professional practice of literacy. These results indicated that it is possible to highlight points deserve highlights and others who deserve to be reconsidered for future training and investigative processes with teaching cases.
O presente estudo se situa nas discussões sobre a formação de professores e investigou a aprendizagem e o desenvolvimento profissional de docentes ao fazerem uso de casos de ensino e de métodos de casos, enquanto ferramenta formativa e investigativa propulsora de processos reflexivos sobre os conhecimentos da docência. Os casos de ensino são narrativas de episódios extraídos geralmente de dados reais do cotidiano escolar e que dão visibilidade às vivências dos professores que, ao analisá los ou construí-los, articulam e revivem experiências, estabelecendo relações entre a teoria e a prática. O processo de pesquisa e intervenção realizou-se por meio do Portal do Professores UFSCar, no curso: Casos de Ensino e Teorização de Práticas Pedagógicas professores alfabetizadores , na perspectiva de formação continuada de dezesseis docentes da mesma etapa do Ensino Fundamental I (Anos Iniciais 1º, 2º e 3º ano), que atuam como alfabetizadoras. Para realização deste estudo, de abordagem qualitativa, a pesquisa teve como objetivo analisar episódios do processo de desenvolvimento profissional vivido por professoras alfabetizadoras ao fazerem uso dos casos de ensino, em ambiente virtual, enquanto possibilidade formativa e investigativa. Para tanto, foram trabalhados casos que retratavam eventos relacionados ao cotidiano das alfabetizadoras. Os resultados alcançados na pesquisa sinalizam a potencialidade dos casos de ensino na/para a aprendizagem e para o desenvolvimento profissional da docência. De modo geral, as estratégias de análise e elaboração de casos de ensino permitiram que conhecimentos referentes ao ensino e aprendizagem da linguagem fossem explicitados e (re)avaliados pelas alfabetizadoras. A utilização de casos de ensino e de métodos de casos permitiu que as reflexões se fizessem presentes e que evidenciassem as dúvidas, as certezas e as contradições que orientam e caracterizam a prática profissional das alfabetizadoras. Com esses resultados apontados é possível evidenciar pontos que merecem destaques e outros que merecem ser repensados para futuros processos formativos e investigativos com casos de ensino.

Le problème “d’apprentissage continu” implique l’entraînement des modèles profonds avec une capacité limitée qui doivent bien fonctionner sur un nombre inconnu de tâches arrivant séquentiellement. Cette configuration peut souvent résulter en un système d’apprentissage qui souffre de “l’oublie catastrophique”, lorsque l’apprentissage d’une nouvelle tâche provoque des interférences sur la progression de l’apprentissage des anciennes tâches. Les travaux récents ont montré que les techniques de “méta-apprentissage” ont le potentiel de ré- duire les interférences entre les anciennes et les nouvelles tâches. Cependant, les procé- dures d’entraînement ont présentement une tendance à être lente ou hors ligne et sensibles à de nombreux hyperparamètres. Dans ce travail, nous proposons “Look-ahead MAML (La-MAML)”, un algorithme de méta-apprentissage rapide basé sur l’optimisation pour l’apprentissage continu en ligne et aidé par une petite mémoire épisodique. Ceci est réalisé en utilisant l’équivalence d’un objectif MAML en plusieurs étapes et un objectif d’apprentissage continu “temps conscient”. L’équivalence résulte au développement d’un algorithme intuitif que nous appelons Continual-MAML (C-MAML), utilisant un méta-apprentissage continu pour optimiser un modèle afin qu’il fonctionne bien sur une série de distributions de don- nées changeantes. En intégrant la modulation des taux d’apprentissage par paramètre dans La-MAML, notre approche fournit un moyen plus flexible et efficace d’atténuer l’oubli catas- trophique par rapport aux méthodes classiques basées sur les prieurs. Cette modulation a également des liens avec des travaux sur la métadescendance, que nous identifions comme une direction importante de la recherche pour développer de meilleurs optimiser pour un ap- prentissage continu. Dans des expériences menées sur des repères de classification visuelle du monde réel, La-MAML atteint des performances supérieures aux autres approches basées sur la relecture, basées sur les prieurs et basées sur le méta-apprentissage pour un apprentissage continu. Nous démontrons également qu’elle est robuste et plus évolutive que de nombreuses approches de pointe.
The continual learning problem involves training models with limited capacity to perform well on a set of an unknown number of sequentially arriving tasks. This setup can of- ten see a learning system undergo catastrophic forgetting, when learning a newly seen task causes interference on the learning progress of old tasks. While recent work has shown that meta-learning has the potential to reduce interference between old and new tasks, the current training procedures tend to be either slow or offline, and sensitive to many hyper-parameters. In this work, we propose Look-ahead MAML (La-MAML), a fast optimisation-based meta- learning algorithm for online-continual learning, aided by a small episodic memory. This is achieved by realising the equivalence of a multi-step MAML objective to a time-aware con- tinual learning objective adopted in prior work. The equivalence leads to the formulation of an intuitive algorithm that we call Continual-MAML (C-MAML), employing continual meta- learning to optimise a model to perform well across a series of changing data distributions. By additionally incorporating the modulation of per-parameter learning rates in La-MAML, our approach provides a more flexible and efficient way to mitigate catastrophic forgetting compared to conventional prior-based methods. This modulation also has connections to prior work on meta-descent, which we identify as an important direction of research to de- velop better optimizers for continual learning. In experiments conducted on real-world visual classification benchmarks, La-MAML achieves performance superior to other replay-based, prior-based and meta-learning based approaches for continual learning. We also demonstrate that it is robust, and more scalable than many recent state-of-the-art approaches.

博士
國立高雄師範大學
工業科技教育學系
101
“Cultivating a global living environment and promoting the general proficiency of the public” are one of the top priorities of the government at present. In Taiwan, not a few local governments have spared no effort establishing English learning environment in middle and elementary schools, creating an all-English context, and implementing a PBL global English village. “Speaking naturally” is incorporated into creative English teaching-learning approach. Through the virtual experiencing in English-context classrooms, the middle and elementary school students’ interest and motives in English learning are aroused, and thus improved. The online learning platform EVEN (English Village Extended Network) is created for a junior high school with an global English village in Pingtung County so that students can have a learning platform with no time-and-space limit, enabling them to keep on learning day-to-day English after they leave the physical environment of global English village. To understand students’ learning efficacy of global English villagers on the further learning website of EVEN, using self-developed teaching and learning questionnaires, we focu on the relationship among their learning motivation, self-adjustment, online cooperation learning, learning environment, learning materials and learning efficacy. The following conclusions are reached: 1.Only a small proportion of students with different variables have conspicuous differences on learning motivation, self-adjustment, online cooperation learning, learning materials and learning efficacy. 2.Tthough the results of confirmatory factor analysis, in terms of standardized factor loadings, accommodation of all levels, and the average variance extracted volume, can not tend to reach the ideal standard as a whole, they are acceptable. Besides, the combined reliability conform to standard value. In all, convergent validity in measurement mode is comparatively good, while its discriminant validity is almost perfect. 3.The whole accommodation of theoretical mode and positive research data constructed from learning motivation, self-adjustment, online cooperation learning, learning materials and learning efficacy are rather satisfactory.

abstract: The rapid advancement of Deep Neural Networks (DNNs), computing, and sensing technology has enabled many new applications, such as the self-driving vehicle, the surveillance drone, and the robotic system. Compared to conventional edge devices (e.g. cell phone or smart home devices), these emerging devices are required to deal with much more complicated and dynamic situations in real-time with bounded computation resources. However, there are several challenges, including but not limited to efficiency, real-time adaptation, model stability, and automation of architecture design. To tackle the challenges mentioned above, model plasticity and stability are leveraged to achieve efficient and online deep learning, especially in the scenario of learning streaming data at the edge: First, a dynamic training scheme named Continuous Growth and Pruning (CGaP) is proposed to compress the DNNs through growing important parameters and pruning unimportant ones, achieving up to 98.1% reduction in the number of parameters. Second, this dissertation presents Progressive Segmented Training (PST), which targets catastrophic forgetting problems in continual learning through importance sampling, model segmentation, and memory-assisted balancing. PST achieves state-of-the-art accuracy with 1.5X FLOPs reduction in the complete inference path. Third, to facilitate online learning in real applications, acquisitive learning (AL) is further proposed to emphasize both knowledge inheritance and acquisition: the majority of the knowledge is first pre-trained in the inherited model and then adapted to acquire new knowledge. The inherited model's stability is monitored by noise injection and the landscape of the loss function, while the acquisition is realized by importance sampling and model segmentation. Compared to a conventional scheme, AL reduces accuracy drop by >10X on CIFAR-100 dataset, with 5X reduction in latency per training image and 150X reduction in training FLOPs. Finally, this dissertation presents evolutionary neural architecture search in light of model stability (ENAS-S). ENAS-S uses a novel fitness score, which addresses not only the accuracy but also the model stability, to search for an optimal inherited model for the application of continual learning. ENAS-S outperforms hand-designed DNNs when learning from a data stream at the edge. In summary, in this dissertation, several algorithms exploiting model plasticity and model stability are presented to improve the efficiency and accuracy of deep neural networks, especially for the scenario of continual learning.
Dissertation/Thesis
Doctoral Dissertation Electrical Engineering 2020

碩士
世新大學
企業管理研究所(含碩專班)
103
Online Game industry is highly competitive，Game companies must use innovation to attract players into the game。Therefore, understanding the behavior of the players will help companies gain a competitive advantage。Learning motivation、Cooperative Learning and Internet addiction is not only a study of education and sociology in very important areas, but also a common phenomenon online gamers 　　Online game will helps player Increased interaction，when player feel interesting，they will Stimulate learning。Players through discussions is an effective way of learning 。According to the literature and discuss the pretest incompatibility issues, improve the reliability and validity of the scale to online gamers for the study, collected questionnaires collected 671 valid questionnaires, response rate was 87.71%. Then use the hypothesis test and regression analysis methods for verification. 　　The results found that (1) there is a significant positive effect on motivation for cooperative learning; (2) cooperative learning has a significant positive impact on the continued intent; (3) Internet Addiction motivation for learning and cooperative learning have a significant moderating effect; (4) Internet addiction intent on cooperation and continuous learning have a significant moderating effect