Dissertations / Theses on the topic 'Attention based models'

Ce travail se situe dans le contexte de la recherche d'information (RI) utilisant des techniques d'intelligence artificielle (IA) telles que l'apprentissage profond (DL). Il s'intéresse à des tâches nécessitant l'appariement de textes, telles que la recherche ad-hoc, le domaine du questions-réponses et l'identification des paraphrases. L'objectif de cette thèse est de proposer de nouveaux modèles, utilisant les méthodes de DL, pour construire des modèles d'appariement basés sur la sémantique de textes, et permettant de pallier les problèmes de l'inadéquation du vocabulaire relatifs aux représentations par sac de mots, ou bag of words (BoW), utilisées dans les modèles classiques de RI. En effet, les méthodes classiques de comparaison de textes sont basées sur la représentation BoW qui considère un texte donné comme un ensemble de mots indépendants. Le processus d'appariement de deux séquences de texte repose sur l'appariement exact entre les mots. La principale limite de cette approche est l'inadéquation du vocabulaire. Ce problème apparaît lorsque les séquences de texte à apparier n'utilisent pas le même vocabulaire, même si leurs sujets sont liés. Par exemple, la requête peut contenir plusieurs mots qui ne sont pas nécessairement utilisés dans les documents de la collection, notamment dans les documents pertinents. Les représentations BoW ignorent plusieurs aspects, tels que la structure du texte et le contexte des mots. Ces caractéristiques sont très importantes et permettent de différencier deux textes utilisant les mêmes mots et dont les informations exprimées sont différentes. Un autre problème dans l'appariement de texte est lié à la longueur des documents. Les parties pertinentes peuvent être réparties de manières différentes dans les documents d'une collection. Ceci est d'autant vrai dans les documents volumineux qui ont tendance à couvrir un grand nombre de sujets et à inclure un vocabulaire variable. Un document long pourrait ainsi comporter plusieurs passages pertinents qu'un modèle d'appariement doit capturer. Contrairement aux documents longs, les documents courts sont susceptibles de concerner un sujet spécifique et ont tendance à contenir un vocabulaire plus restreint. L'évaluation de leur pertinence est en principe plus simple que celle des documents plus longs. Dans cette thèse, nous avons proposé différentes contributions répondant chacune à l'un des problèmes susmentionnés. Tout d'abord, afin de résoudre le problème d'inadéquation du vocabulaire, nous avons utilisé des représentations distribuées des mots (plongement lexical) pour permettre un appariement basé sur la sémantique entre les différents mots. Ces représentations ont été utilisées dans des applications de RI où la similarité document-requête est calculée en comparant tous les vecteurs de termes de la requête avec tous les vecteurs de termes du document, indifféremment. Contrairement aux modèles proposés dans l'état-de-l'art, nous avons étudié l'impact des termes de la requête concernant leur présence/absence dans un document. Nous avons adopté différentes stratégies d'appariement document/requête. L'intuition est que l'absence des termes de la requête dans les documents pertinents est en soi un aspect utile à prendre en compte dans le processus de comparaison. En effet, ces termes n'apparaissent pas dans les documents de la collection pour deux raisons possibles : soit leurs synonymes ont été utilisés ; soit ils ne font pas partie du contexte des documents en questions<br>This work is situated in the context of information retrieval (IR) using machine learning (ML) and deep learning (DL) techniques. It concerns different tasks requiring text matching, such as ad-hoc research, question answering and paraphrase identification. The objective of this thesis is to propose new approaches, using DL methods, to construct semantic-based models for text matching, and to overcome the problems of vocabulary mismatch related to the classical bag of word (BoW) representations used in traditional IR models. Indeed, traditional text matching methods are based on the BoW representation, which considers a given text as a set of independent words. The process of matching two sequences of text is based on the exact matching between words. The main limitation of this approach is related to the vocabulary mismatch. This problem occurs when the text sequences to be matched do not use the same vocabulary, even if their subjects are related. For example, the query may contain several words that are not necessarily used in the documents of the collection, including relevant documents. BoW representations ignore several aspects about a text sequence, such as the structure the context of words. These characteristics are important and make it possible to differentiate between two texts that use the same words but expressing different information. Another problem in text matching is related to the length of documents. The relevant parts can be distributed in different ways in the documents of a collection. This is especially true in large documents that tend to cover a large number of topics and include variable vocabulary. A long document could thus contain several relevant passages that a matching model must capture. Unlike long documents, short documents are likely to be relevant to a specific subject and tend to contain a more restricted vocabulary. Assessing their relevance is in principle simpler than assessing the one of longer documents. In this thesis, we have proposed different contributions, each addressing one of the above-mentioned issues. First, in order to solve the problem of vocabulary mismatch, we used distributed representations of words (word embedding) to allow a semantic matching between the different words. These representations have been used in IR applications where document/query similarity is computed by comparing all the term vectors of the query with all the term vectors of the document, regardless. Unlike the models proposed in the state-of-the-art, we studied the impact of query terms regarding their presence/absence in a document. We have adopted different document/query matching strategies. The intuition is that the absence of the query terms in the relevant documents is in itself a useful aspect to be taken into account in the matching process. Indeed, these terms do not appear in documents of the collection for two possible reasons: either their synonyms have been used or they are not part of the context of the considered documents. The methods we have proposed make it possible, on the one hand, to perform an inaccurate matching between the document and the query, and on the other hand, to evaluate the impact of the different terms of a query in the matching process. Although the use of word embedding allows semantic-based matching between different text sequences, these representations combined with classical matching models still consider the text as a list of independent elements (bag of vectors instead of bag of words). However, the structure of the text as well as the order of the words is important. Any change in the structure of the text and/or the order of words alters the information expressed. In order to solve this problem, neural models were used in text matching

The main focus of this thesis is to develop biologically-based computational models for object recognition. A series of models for attention and object recognition were developed in the order of increasing functionality and complexity. These models are based on information processing in the primate brain, and specially inspired from the theory of visual information processing along the two parallel processing pathways of the primate visual cortex. To capture the true essence of incremental, constraint satisfaction style processing in the visual system, interactive neural networks were used for implementing our models. Results from eye-tracking studies on the relevant visual tasks, as well as our hypothesis regarding the information processing in the primate visual system, were implemented in the models and tested with simulations. As a first step, a model based on the ventral pathway was developed to recognize single objects. Through systematic testing, structural and algorithmic parameters of these models were fine tuned for performing their task optimally. In the second step, the model was extended by considering the dorsal pathway, which enables simulation of visual attention as an emergent phenomenon. The extended model was then investigated for visual search tasks. In the last step, we focussed on occluded and overlapped object recognition. A couple of eye-tracking studies were conducted in this regard and on the basis of the results we made some hypotheses regarding information processing in the primate visual system. The models were further advanced on the lines of the presented hypothesis, and simulated on the tasks of occluded and overlapped object recognition. On the basis of the results and analysis of our simulations we have further found that the generalization performance of interactive hierarchical networks improves with the addition of a small amount of Hebbian learning to an otherwise pure error-driven learning. We also concluded that the size of the receptive fields in our networks is an important parameter for the generalization task and depends on the object of interest in the image. Our results show that networks using hard coded feature extraction perform better than the networks that use Hebbian learning for developing feature detectors. We have successfully demonstrated the emergence of visual attention within an interactive network and also the role of context in the search task. Simulation results with occluded and overlapped objects support our extended interactive processing approach, which is a combination of the interactive and top-down approach, to the segmentation-recognition issue. Furthermore, the simulation behavior of our models is in line with known human behavior for similar tasks. In general, the work in this thesis will improve the understanding and performance of biologically-based interactive networks for object recognition and provide a biologically-plausible solution to recognition of occluded and overlapped objects. Moreover, our models provide some suggestions for the underlying neural mechanism and strategies behind biological object recognition.

UOL; CAPES<br>Esta tese apresenta um método para a extração de regiões de interesse (ROIs) de imagens. No contexto deste trabalho, ROIs são definidas como os objetos semânticos que se destacam em uma imagem, podendo apresentar qualquer tamanho ou localização. O novo método baseia-se em modelos computacionais de atenção visual (VA), opera de forma completamente bottom-up, não supervisionada e não apresenta restrições com relação à categoria da imagem de entrada. Os elementos centrais da arquitetura são os modelos de VA propostos por Itti-Koch-Niebur e Stentiford. O modelo de Itti-Koch-Niebur considera as características de cor, intensidade e orientação da imagem e apresenta uma resposta na forma de coordenadas, correspondentes aos pontos de atenção (POAs) da imagem. O modelo Stentiford considera apenas as características de cor e apresenta a resposta na forma de áreas de atenção na imagem (AOAs). Na arquitetura proposta, a combinação de POAs e AOAs permite a obtenção dos contornos das ROIs. Duas implementações desta arquitetura, denominadas 'primeira versão' e 'versão melhorada' são apresentadas. A primeira versão utiliza principalmente operações tradicionais de morfologia matemática. Esta versão foi aplicada em dois sistemas de recuperação de imagens com base em regiões. No primeiro, as imagens são agrupadas de acordo com as ROIs, ao invés das características globais da imagem. O resultado são grupos de imagens mais significativos semanticamente, uma vez que o critério utilizado são os objetos da mesma categoria contidos nas imagens. No segundo sistema, á apresentada uma combinação da busca de imagens tradicional, baseada nas características globais da imagem, com a busca de imagens baseada em regiões. Ainda neste sistema, as buscas são especificadas através de mais de uma imagem exemplo. Na versão melhorada da arquitetura, os estágios principais são uma análise de coerência espacial entre as representações de ambos modelos de VA e uma representação multi-escala das AOAs. Se comparada à primeira versão, esta apresenta maior versatilidade, especialmente com relação aos tamanhos das ROIs presentes nas imagens. A versão melhorada foi avaliada diretamente, com uma ampla variedade de imagens diferentes bancos de imagens públicos, com padrões-ouro na forma de bounding boxes e de contornos reais dos objetos. As métricas utilizadas na avaliação foram presision, recall, F1 e area of overlap. Os resultados finais são excelentes, considerando-se a abordagem exclusivamente bottom-up e não-supervisionada do método.<br>This thesis presents a method for the extraction of regions of interest (ROIs) from images. By ROIs we mean the most prominent semantic objects in the images, of any size and located at any position in the image. The novel method is based on computational models of visual attention (VA), operates under a completely bottom-up and unsupervised way and does not present con-straints in the category of the input images. At the core of the architecture is de model VA proposed by Itti, Koch and Niebur and the one proposed by Stentiford. The first model takes into account color, intensity, and orientation features and provides coordinates corresponding to the points of attention (POAs) in the image. The second model considers color features and provides rough areas of attention (AOAs) in the image. In the proposed architecture, the POAs and AOAs are combined to establish the contours of the ROIs. Two implementations of this architecture are presented, namely 'first version' and 'improved version'. The first version mainly on traditional morphological operations and was applied in two novel region-based image retrieval systems. In the first one, images are clustered on the basis of the ROIs, instead of the global characteristics of the image. This provides a meaningful organization of the database images, since the output clusters tend to contain objects belonging to the same category. In the second system, we present a combination of the traditional global-based with region-based image retrieval under a multiple-example query scheme. In the improved version of the architecture, the main stages are a spatial coherence analysis between both VA models and a multiscale representation of the AOAs. Comparing to the first one, the improved version presents more versatility, mainly in terms of the size of the extracted ROIs. The improved version was directly evaluated for a wide variety of images from different publicly available databases, with ground truth in the form of bounding boxes and true object contours. The performance measures used were precision, recall, F1 and area overlap. Experimental results are of very high quality, particularly if one takes into account the bottom-up and unsupervised nature of the approach.

Linear mixed models (LMMs) provide a still underused methodological perspective on combining experimental and individual-differences research. Here we illustrate this approach with two-rectangle cueing in visual attention (Egly et al., 1994). We replicated previous experimental cue-validity effects relating to a spatial shift of attention within an object (spatial effect), to attention switch between objects (object effect), and to the attraction of attention toward the display centroid (attraction effect), also taking into account the design-inherent imbalance of valid and other trials. We simultaneously estimated variance/covariance components of subject-related random effects for these spatial, object, and attraction effects in addition to their mean reaction times (RTs). The spatial effect showed a strong positive correlation with mean RT and a strong negative correlation with the attraction effect. The analysis of individual differences suggests that slow subjects engage attention more strongly at the cued location than fast subjects. We compare this joint LMM analysis of experimental effects and associated subject-related variances and correlations with two frequently used alternative statistical procedures

With the recent advantages of machine learning in cheminformatics, the drug discovery process has been accelerated; providing a high impact in the field of medicine and public health. Molecular property and activity prediction are key elements in the early stages of drug discovery by helping prioritize the experiments and reduce the experimental work. In this thesis, a novel approach for multi-task regression using a text-based Transformer model is introduced and thoroughly explored for training on a number of properties or activities simultaneously. This multi-task regression with Transformer based model is inspired by the field of Natural Language Processing (NLP) which uses prefix tokens to distinguish between each task. In order to investigate our architecture two data categories are used; 133 biological activities from ExCAPE database and three physical chemistry properties from MoleculeNet benchmark datasets. The Transformer model consists of the embedding layer with positional encoding, a number of encoder layers, and a Feedforward Neural Network (FNN) to turn it into a regression problem. The molecules are represented as a string of characters using the Simplified Molecular-Input Line-Entry System (SMILES) which is a ’chemistry language’ with its own syntax. In addition, the effect of Transfer Learning is explored by experimenting with two pretrained Transformer models, pretrained on 1.5 million and on 100 million molecules. The text-base Transformer models are compared with a feature-based Support Vector Regression (SVR) with the Tanimoto kernel where the input molecules are encoded as Extended Connectivity Fingerprint (ECFP), which are calculated features. The results have shown that Transfer Learning is crucial for improving the performance on both property and activity predictions. On bioactivity tasks, the larger pretrained Transformer on 100 million molecules achieved comparable performance to the feature-based SVR model; however, overall SVR performed better on the majority of the bioactivity tasks. On the other hand, on physicochemistry property tasks, the larger pretrained Transformer outperformed SVR on all three tasks. Concluding, the multi-task regression architecture with the prefix token had comparable performance with the traditional feature-based approach on predicting different molecular properties or activities. Lastly, using the larger pretrained models trained on a wide chemical space can play a key role in improving the performance of Transformer models on these tasks.

The availability and use of knowledge graphs have become commonplace as a compact storage of information and for lookup of facts. However, the discrete representation makes the knowledge graph unavailable for tasks that need a continuous representation, such as predicting relationships between entities, where the most probable relationship needs to be found. The need for a continuous representation has spurred the development of knowledge graph embeddings. The idea is to position the entities of the graph relative to each other in a continuous low-dimensional vector space, so that their relationships are preserved, and ideally leading to clusters of entities with similar characteristics. Several methods to produce knowledge graph embeddings have been created, from simple models that minimize the distance between related entities to complex neural models. Almost all of these embedding methods attempt to create an accurate static representation of each entity and relation. However, as with words in natural language, both entities and relations in a knowledge graph hold different meanings in different local contexts.  With the recent development of Transformer models, and their success in creating contextual representations of natural language, work has been done to apply them to graphs. Initial results show great promise, but there are significant differences in archi- tecture design across papers. There is no clear direction on how Transformer models can be best applied to create contextual knowledge graph embeddings. Two of the main differences in previous work is how the attention mask is applied in the model and what input graph structures the model is trained on.  This report explores how different attention masking methods and graph inputs affect a Transformer model (in this report, BERT) on a link prediction task for triples. Models are trained with five different attention masking methods, which to varying degrees restrict attention, and on three different input graph structures (triples, paths, and interconnected triples).  The results indicate that a Transformer model trained with a masked language model objective has the strongest performance on the link prediction task when there are no restrictions on how attention is directed, and when it is trained on graph structures that are sequential. This is similar to how models like BERT learn sentence structure after being exposed to a large number of training samples. For more complex graph structures it is beneficial to encode information of the graph structure through how the attention mask is applied. There also seems to be some indications that the input graph structure affects the models’ capabilities to learn underlying characteristics in the knowledge graph that is trained upon.

Im ersten Teil analysiere ich die Selbstorganisation zur Kritikalität (hier ein Phasenübergang von Ordnung zu Unordnung) und untersuche, ob Evolution ein möglicher Organisationsmechanismus ist. Die Kernfrage ist, ob sich ein simulierter kohäsiver Schwarm, der versucht, einem Raubtier auszuweichen, durch Evolution selbst zum kritischen Punkt entwickelt, um das Ausweichen zu optimieren? Es stellt sich heraus, dass (i) die Gruppe den Jäger am besten am kritischen Punkt vermeidet, aber (ii) nicht durch einer verstärkten Reaktion, sondern durch strukturelle Veränderungen, (iii) das Gruppenoptimum ist evolutionär unstabiler aufgrund einer maximalen räumlichen Selbstsortierung der Individuen. Im zweiten Teil modelliere ich experimentell beobachtete Unterschiede im kollektiven Verhalten von Fischgruppen, die über mehrere Generationen verschiedenen Arten von größenabhängiger Selektion ausgesetzt waren. Diese Größenselektion soll Freizeitfischerei (kleine Fische werden freigelassen, große werden konsumiert) und die kommerzielle Fischerei mit großen Netzbreiten (kleine/junge Individuen können entkommen) nachahmen. Die zeigt sich, dass das Fangen großer Fische den Zusammenhalt und die Risikobereitschaft der Individuen reduziert. Beide Befunde lassen sich mechanistisch durch einen Aufmerksamkeits-Kompromiss zwischen Sozial- und Umweltinformationen erklären. Im letzten Teil der Arbeit quantifiziere ich die kollektive Informationsverarbeitung im Feld. Das Studiensystem ist eine an sulfidische Wasserbedingungen angepasste Fischart mit einem kollektiven Fluchtverhalten vor Vögeln (wiederholte kollektive Fluchttauchgängen). Die Fische sind etwa 2 Zentimeter groß, aber die kollektive Welle breitet sich über Meter in dichten Schwärmen an der Oberfläche aus. Es zeigt sich, dass die Wellengeschwindigkeit schwach mit der Polarisation zunimmt, bei einer optimalen Dichte am schnellsten ist und von ihrer Richtung relativ zur Schwarmorientierung abhängt.<br>In the first part, I focus on the self-organization to criticality (here an order-disorder phase transition) and investigate if evolution is a possible self-tuning mechanism. Does a simulated cohesive swarm that tries to avoid a pursuing predator self-tunes itself by evolution to the critical point to optimize avoidance? It turns out that (i) the best group avoidance is at criticality but (ii) not due to an enhanced response but because of structural changes (fundamentally linked to criticality), (iii) the group optimum is not an evolutionary stable state, in fact (iv) it is an evolutionary accelerator due to a maximal spatial self-sorting of individuals causing spatial selection. In the second part, I model experimentally observed differences in collective behavior of fish groups subject to multiple generation of different types of size-dependent selection. The real world analog to this experimental evolution is recreational fishery (small fish are released, large are consumed) and commercial fishing with large net widths (small/young individuals can escape). The results suggest that large harvesting reduces cohesion and risk taking of individuals. I show that both findings can be mechanistically explained based on an attention trade-off between social and environmental information. Furthermore, I numerically analyze how differently size-harvested groups perform in a natural predator and fishing scenario. In the last part of the thesis, I quantify the collective information processing in the field. The study system is a fish species adapted to sulfidic water conditions with a collective escape behavior from aerial predators which manifests in repeated collective escape dives. These fish measure about 2 centimeters, but the collective wave spreads across meters in dense shoals at the surface. I find that wave speed increases weakly with polarization, is fastest at an optimal density and depends on its direction relative to shoal orientation.

This study focused on policy agenda-setting models for school sport and physical education in South Africa. The primary objective was to assess and propose options for improved agenda-setting by focussing on the use of agenda-setting models and by applying it to physical education and school sport and the policy agenda of the national government. The study has shown that pertinent school sport and physical education policy issues, as supported by key role-players and principal actors, were initially not placed on the formal policy agenda of government during the research investigation period (2005-2009). However, during 2010 and 2011 the issue of school sport and physical education received prominent attention by authorities and these developments were subsequently included in the study. The study aimed at contributing to existing policy agenda-setting models and by recommending changes to the Generic Process Model.The study also made a contribution by informing various role-players and stakeholders in education and school sport on the opportunities in policy agenda-setting. The study showed that policy agenda-setting is a vital step in the Generic Policy Process Model. Policy agendasetting in South Africa is critical, as it is important to place new and emerging policy issues on the policy agenda and as a participative public policy process is relatively new in this young democracy. The reader should not confuse the study as one dealing with school sport and physical education primarily, but rather as a research investigation dealing with policy agenda-setting models as applied to school sport and physical education.The secondary objectives of the study included the development of a historical perspective on trends and tendencies in education and sport in South Africa. A second objective was to provide theoretical perspectives on public policy and specifically on policy agenda-setting. From these theoretical perspectives, the Generic Policy Process Model was selected to use as a model that provided guidance on the overall policy process normally followed in South Africa. The Issue Attention Cycle and Principal Actor Models on Agenda-Setting were selected to apply to the case study to specifically ascertain important factors related to policy agenda-setting such as the identification of key role players as well as key policy issues. The Generic Policy Process Model provided for both a comprehensive set of phases as well as specific requirements and key issues to be addressed during each phase of the policy process.In terms of findings the study found that a number of specific agenda-setting elements or phases needed to be added to the Generic Policy Process Model, which includes a problem stage, triggers, initiator, issue creation and actors or policy stakeholders.The Principal Actor Model to agenda-setting was selected for application to the case as different actors have different levels of success at each policy stage. In the South African experience it is important to look at who sets the policy agenda and why, who can initiate agenda-setting and the role played by these principal actors in the agenda-setting process.Issue emergence often places policy issues on the policy agenda. The public is initially involved in issues, but in the long term public interest declines. The government realizes the significant costs involved in placing policy issues back on the agenda. This leads to a decline in issue attention by policy-makers and the public. The Issue Attention Cycle Model of agenda setting was used to analyse this phenomenon in South African Education policy.The study provides a case assessment of the South African experience. From the research findings, a set of conclusions and recommendations were developed for improved policy agenda-setting models and implications for school sport and physical education, as well as tools to place it on the national policy agenda were identified. The research findings suggest that pertinent school sport and physical education policy issues, as supported by key roleplayers,stakeholders and principal actors were not placed on the formal policy agenda of the government as a vital step in the policy process between 2005 and 2009. Ever since, principal policy actors, civil society NGOs, and government officials placed sufficient pressure on the Minister of Basic Education to place Physical Education on the agenda. Subsequently,Minister Angie Motshega has placed physical education in the school Curriculum under the subject Life Orientation and Lifeskills. It has become evident from the research that agendasetting is both necessary to, and a complex phase in, the policy-making process.This study has shown that major policy issues such as physical education and school sport were neglected during the period 2005 and 2009 despite reformed and advanced policy cycles in government. It has also shown that the role of policy agenda-setting in the overall policymaking process was revisited by government in the subsequent period 2010/2011 and placed on the policy agenda. Specific lessons of experience emanated from this process.The study recommends that the triggers of the agenda-setting phases be added to the Generic Policy Process Model, which should include the problem stage, triggers, initiators, issue creation, actors and policy stakeholders. Principal actors in the agenda-setting model in South Africa want the issue of physical education and school sport to be part of the school curriculum, and therefore be placed back on the policy agenda by the Government on its institutional agenda. Furthermore, the study showed that actors wanted it to be compulsory in all phases of the school (Foundation, Intermediate, Senior, GET, FET) and that it should have the same legal status as other subjects.The important findings include that: Comprehensive policy process models such as that of Dunn, Wissink and the Generic Process model may need to be reviewed to incorporate more fully the policy-agenda setting stages of the overall process; Current policy agenda setting models in use are relevant and valuable in identifying key role players as well as key issues and considerations regarding the policy process; Institutional arrangements to strengthen the role of NGOs and lower level institutions,such as schools to participate in policy agenda setting are important; and the study has shown that a number of key factors have been identified that had a key influence on policy agenda-setting in the case of physical education and school sport in South Africa. These included the influence of changing political leadership, the competency of policy capacities in government, the profile of issues in the media etc. The key findings of the study have shown that further potential exists to improve monitoring and evaluation and policy analysis.The study made a set of recommendations to principal actors such as the Minister of Education, Minister of Sport and Recreation, non-governmental organisations, interest groups,department officials and pressure groups. A set of research topics was also identified for future research.<br>Philosophiae Doctor - PhD

This work concerns the investigation of the neuronal mechanisms at the basis of language acquisition and processing, and the complex interactions of language and attention processes in the human brain. In particular, this research was motivated by two sets of existing neurophysiological data which cannot be reconciled on the basis of current psycholinguistic accounts: on the one hand, the N400, a robust index of lexico-semantic processing which emerges at around 400ms after stimulus onset in attention demanding tasks and is larger for senseless materials (meaningless pseudowords) than for matched meaningful stimuli (words); on the other, the more recent results on the Mismatch Negativity (MMN, latency 100-250ms), an early automatic brain response elicited under distraction which is larger to words than to pseudowords. We asked what the mechanisms underlying these differential neurophysiological responses may be, and whether attention and language processes could interact so as to produce the observed brain responses, having opposite magnitude and different latencies. We also asked questions about the functional nature and anatomical characteristics of the cortical representation of linguistic elements. These questions were addressed by combining neurocomputational techniques and neuroimaging (magneto-encephalography, MEG) experimental methods. Firstly, a neurobiologically realistic neural-network model composed of neuron-like elements (graded response units) was implemented, which closely replicates the neuroanatomical and connectivity features of the main areas of the left perisylvian cortex involved in spoken language processing (i.e., the areas controlling speech output – left inferior-prefrontal cortex, including Broca’s area – and the main sensory input – auditory – areas, located in the left superior-temporal lobe, including Wernicke’s area). Secondly, the model was used to simulate early word acquisition processes by means of a Hebbian correlation learning rule (which reflects known synaptic plasticity mechanisms of the neocortex). The network was “taught” to associate pairs of auditory and articulatory activation patterns, simulating activity due to perception and production of the same speech sound: as a result, neuronal word representations distributed over the different cortical areas of the model emerged. Thirdly, the network was stimulated, in its “auditory cortex”, with either one of the words it had learned, or new, unfamiliar pseudoword patterns, while the availability of attentional resources was modulated by changing the level of non-specific, global cortical inhibition. In this way, the model was able to replicate both the MMN and N400 brain responses by means of a single set of neuroscientifically grounded principles, providing the first mechanistic account, at the cortical-circuit level, for these data. Finally, in order to verify the neurophysiological validity of the model, its crucial predictions were tested in a novel MEG experiment investigating how attention processes modulate event-related brain responses to speech stimuli. Neurophysiological responses to the same words and pseudowords were recorded while the same subjects were asked to attend to the spoken input or ignore it. The experimental results confirmed the model’s predictions; in particular, profound variability of magnetic brain responses to pseudowords but relative stability of activation to words as a function of attention emerged. While the results of the simulations demonstrated that distributed cortical representations for words can spontaneously emerge in the cortex as a result of neuroanatomical structure and synaptic plasticity, the experimental results confirm the validity of the model and provide evidence in support of the existence of such memory circuits in the brain. This work is a first step towards a mechanistic account of cognition in which the basic atoms of cognitive processing (e.g., words, objects, faces) are represented in the brain as discrete and distributed action-perception networks that behave as closed, independent systems.

Pour les robots mobiles autonomes conçus pour partager notre environnement, la sécurité et l'efficacité de leur trajectoire ne sont pas les seuls aspects à prendre en compte pour la planification de leur mouvement: ils doivent respecter des règles sociales afin de ne pas gêner les personnes environnantes. Dans un tel contexte social, la plupart des techniques de planification de mouvement actuelles s'appuient fortement sur le concept d'espaces sociaux; de tels espaces sociaux sont cependant difficiles à modéliser et ils sont d'une utilisation limitée dans le contexte d'interactions homme-robot où l'intrusion dans les espaces sociaux est nécessaire. Ce travail présente une nouvelle approche pour la planification de mouvements dans un contexte social qui permet de gérer des environnements complexes ainsi que des situation d’interaction homme-robot. Plus précisément, le concept d'attention est utilisé pour modéliser comment l'influence de l'environnement dans son ensemble affecte la manière dont le mouvement du robot est perçu par les personnes environnantes. Un nouveau modèle attentionnel est introduit qui estime comment nos ressources attentionnelles sont partagées entre les éléments saillants de notre environnement. Basé sur ce modèle, nous introduisons le concept de champ attentionnel. Un planificateur de mouvement est ensuite développé qui s'appuie sur le champ attentionnel afin de produire des mouvements socialement acceptables. Notre planificateur de mouvement est capable d'optimiser simultanément plusieurs objectifs tels que la sécurité, l'efficacité et le confort des mouvements. Les capacités de l'approche proposée sont illustrées sur plusieurs scénarios simulés dans lesquels le robot est assigné différentes tâches. Lorsque la tâche du robot consiste à naviguer dans l'environnement sans causer de distraction, notre approche produit des résultats prometteurs même dans des situations complexes. Aussi, lorsque la tâche consiste à attirer l'attention d'une personne en vue d'interagir avec elle, notre planificateur de mouvement est capable de choisir automatiquement une destination qui exprime au mieux son désir d'interagir, tout en produisant un mouvement sûr, efficace et confortable<br>For autonomous mobile robots designed to share their environment with humans, path safety and efficiency are not the only aspects guiding their motion: they must follow social rules so as not to cause discomfort to surrounding people. Most socially-aware path planners rely heavily on the concept of social spaces; however, social spaces are hard to model and they are of limited use in the context of human-robot interaction where intrusion into social spaces is necessary. In this work, a new approach for socially-aware path planning is presented that performs well in complex environments as well as in the context of human-robot interaction. Specifically, the concept of attention is used to model how the influence of the environment as a whole affects how the robot's motion is perceived by people within close proximity. A new computational model of attention is presented that estimates how our attentional resources are shared amongst the salient elements in our environment. Based on this model, the novel concept of attention field is introduced and a path planner that relies on this field is developed in order to produce socially acceptable paths. To do so, a state-of-the-art many-objective optimization algorithm is successfully applied to the path planning problem. The capacities of the proposed approach are illustrated in several case studies where the robot is assigned different tasks. Firstly, when the task is to navigate in the environment without causing distraction our approach produces promising results even in complex situations. Secondly, when the task is to attract a person's attention in view of interacting with him or her, the motion planner is able to automatically choose a destination that best conveys its desire to interact whilst keeping the motion safe, efficient and socially acceptable

Modern computer graphics systems are able to construct renderings of such high quality that viewers are deceived into regarding the images as coming from a photographic source. Large amounts of computing resources are expended in this rendering process, using complex mathematical models of lighting and shading. However, psychophysical experiments have revealed that viewers only regard certain informative regions within a presented image. Furthermore, it has been shown that these visually important regions contain low-level visual feature differences that attract the attention of the viewer. This thesis will present a new approach to image synthesis that exploits these experimental findings by modulating the spatial quality of image regions by their visual importance. Efficiency gains are therefore reaped, without sacrificing much of the perceived quality of the image. Two tasks must be undertaken to achieve this goal. Firstly, the design of an appropriate region-based model of visual importance, and secondly, the modification of progressive rendering techniques to effect an importance-based rendering approach. A rule-based fuzzy logic model is presented that computes, using spatial feature differences, the relative visual importance of regions in an image. This model improves upon previous work by incorporating threshold effects induced by global feature difference distributions and by using texture concentration measures. A modified approach to progressive ray-tracing is also presented. This new approach uses the visual importance model to guide the progressive refinement of an image. In addition, this concept of visual importance has been incorporated into supersampling, texture mapping and computer animation techniques. Experimental results are presented, illustrating the efficiency gains reaped from using this method of progressive rendering. This visual importance-based rendering approach is expected to have applications in the entertainment industry, where image fidelity may be sacrificed for efficiency purposes, as long as the overall visual impression of the scene is maintained. Different aspects of the approach should find many other applications in image compression, image retrieval, progressive data transmission and active robotic vision.

A computational cognitive neuroscience approach was used to examine processes of visual attention in the human and monkey brain. The aim of the work was to produce a biologically plausible neurodynamical model of both spatial and object-based attention that accounted for observations in monkey visual areas V4, inferior temporal cortex (IT) and the lateral intraparietal area (LIP), and was able to produce search scan path behaviour similar to that observed in humans and monkeys. Of particular interest currently in the visual attention literature is the biased competition hypothesis (Desimone &amp; Duncan. 1995). The model presented here is the first active vision implementation of biased competition, where attcntional shifts are overt. Therefore, retinal inputs change during the scan path and this approach raised issues, such as memory for searched locations across saccades, not addressed bv previous models with static retinas. This is the first model to examine the different time courses associated with spatial and object-based effects at the cellular level. Single cell recordings in areas V4 (Luck et al., 1997; Chelazzi et al., 2001) and IT (Chelazzi ct al., 1993, 1998) were replicated such that attentional effects occurred at the appropriate time after onset of the stimulus. Object-based effects at the cellular level of the model led to systems level behaviour that replicated that observed during active visual search for orientation and colour feature conjunction targets in psychophysical investigations. This provides a valuable insight into the link between cellular and system level behaviour in natural systems. At the systems level, the simulated search process showed selectivity in its scan path that was similar to that observed in humans (Scialfa &amp; Joffe, 1998; Williams &amp; Reingold, 2001) and monkeys (Motter &amp; Belky. 1998b), being guided to target coloured locations in preference to locations containing the target orientation or blank areas. A connection between the ventral and dorsal visual processing streams (Ungerleider &amp; Mishkin. 1982) is suggested to contribute to this selectivity and priority in the featural guidance of search. Such selectivity and avoidance of blank areas has potential application in computer vision applications. Simulation of lesions within the model and comparison with patient data provided further verification of the model. Simulation of visual neglect due to parietal cortical lesion suggests that the model has the capability to provide insights into the neural correlates of the conscious perception of stimuli The biased competition approach described here provides an extendable framework within which further &quot;bottom-up&quot; stimulus and &quot;top-down&quot; mnemonic and cognitive biases can be added, in order to further examine exogenous versus endogenous factors in the capture of attention.

Visual attention can be deployed to locations within the visual array (spatial attention), to individual features such as colour and form (feature-based attention), or to entire objects (object-based attention). Objects are composed of features to form a perceived ‘whole’. This compositional object representation reduces the storage demands by avoiding the need to store every type of object experienced. However, this approach exposes a problem of binding these constituent features (e.g. form and colour) into objects. The problem is made explicit in the higher areas of the ventral stream as information about a feature’s location is absent. For feature-based attention and search, activations flow from the inferotemporal cortex to primary visual cortex without spatial cues from the dorsal stream, therefore the neural effect is applied to all locations across the visual field [79, 60, 7, 52]. My research hypothesis is that biased competition occurs independently for each cued feature, and is implemented by lateral inhibition between a feedforward and a feedback network through a cortical micro-circuit architecture. The local competition for each feature can be combined in the dorsal stream via spatial congruence to implement a secondary spatial attention mechanism, and in early visual areas to bind together the distributed featural representation of a target object.

Using the attention-based view, this study is concerned with two levels of board of directors’ interaction relating to strategic entrepreneurship: (1) how individual board members may affect the attention of the entire board, and (2) how the board may affect the attention and resource allocation of the firm. Unique to prior literature, this study considers contextual factors at each level of interaction and views the board room communications through unprecedented access. Multiple regression and negative binomial regression analyses are used to test the theoretical hypotheses.

With the increasing highlighted security concerns in Intelligent Transportation System (ITS), Vision-based Automated Vehicle Recognition (VAVR) has attracted considerable attention recently. A comprehensive VAVR system contains three components: Vehicle Detection (VD), Vehicle Make and Model Recognition (VMMR), and Vehicle Re-identification (VReID). These components perform coarse-to-fine recognition tasks in three steps. The VAVR system can be widely used in suspicious vehicle recognition, urban traffic monitoring, and automated driving system. Vehicle recognition is complicated due to the subtle visual differences between different vehicle models. Therefore, how to build a VAVR system that can fast and accurately recognize vehicle information has gained tremendous attention. In this work, by taking advantage of the emerging deep learning methods, which have powerful feature extraction and pattern learning abilities, we propose several models used for vehicle recognition. First, we propose a novel Recurrent Attention Unit (RAU) to expand the standard Convolutional Neural Network (CNN) architecture for VMMR. RAU learns to recognize the discriminative part of a vehicle on multiple scales and builds up a connection with the prominent information in a recurrent way. The proposed ResNet101-RAU achieves excellent recognition accuracy of 93.81% on the Stanford Cars dataset and 97.84% on the CompCars dataset. Second, to construct efficient vehicle recognition models, we simplify the structure of RAU and propose a Lightweight Recurrent Attention Unit (LRAU). The proposed LRAU extracts the discriminative part features by generating attention masks to locate the keypoints of a vehicle (e.g., logo, headlight). The attention mask is generated based on the feature maps received by the LRAU and the preceding attention state generated by the preceding LRAU. Then, by adding LRAUs to the standard CNN architectures, we construct three efficient VMMR models. Our models achieve the state-of-the-art results with 93.94% accuracy on the Stanford Cars dataset, 98.31% accuracy on the CompCars dataset, and 99.41% on the NTOU-MMR dataset. In addition, we construct a one-stage Vehicle Detection and Fine-grained Recognition (VDFG) model by combining our LRAU with the general object detection model. Results show the proposed VDFG model can achieve excellent performance with real-time processing speed. Third, to address the VReID task, we design the Compact Attention Unit (CAU). CAU has a compact structure, and it relies on a single attention map to extract the discriminative local features of a vehicle. We add two CAUs to the truncated ResNet to construct a small but efficient VReID model, ResNetT-CAU. Compared with the original ResNet, the model size of ResNetT-CAU is reduced by 60%. Extensive experiments on the VeRi and VehicleID dataset indicate the proposed ResNetT-CAU achieve the best re-identification results on both datasets. In summary, the experimental results on the challenging benchmark VMMR and VReID datasets indicate our models achieve the best VMMR and VReID performance, and our models have a small model size and fast image processing speed.

碩士<br>國立中央大學<br>資訊工程學系<br>107<br>Emotional Intelligence is a field from which awareness is heavily being raised. Coupled with language generation, one expects to further humanize the machine and be a step closer to the user by generating responses that are consistent with a specific emotion. The analysis of sentiment within documents or sentences have been widely studied and improved while the generation of emotional content remains under-researched. Meanwhile, generative models have recently known series of improvements thanks to Generative Adversarial Network (GAN). Promising results are frequently reported in both natural language processing and computer vision. However, when applied to text generation, adversarial learning may lead to poor quality sentences and mode collapse. In this paper, we leverage one-round data conversation from social media to propose a novel approach in order to generate grammatically-correct-and-emotional-consistent answers for Short-Text Conversation task (STC-3) for NTCIR-14 workshop. We make use of an Attention-based Sequence-to-Sequence as our generator, inspired from StarGAN framework. We provide emotion embeddings and direct feedback from an emotion classifier to guide the generator. To avoid the aforementioned issues with adversarial networks, we alternatively train our generator using maximum likelihood and adversarial loss.

The intention of the project is to determine whether political elites have to discuss an issue using a specific emotional tone before the public and other political elites consider that issue a problem. Research has not yet demonstrated under what conditions elite rhetorical cues can heighten issue attention. Past studies have suggested that an increase in the absolute intensity of elite issue discussion can heighten perceptions of an issue as a problem. The problem with this notion is that within that absolute issue discussion, elites might simply be repeatedly saying conditions related to an issue are stable. They might also be presenting basic factual background information about an issue, a type of discussion unlikely to capture the interest of many in the political system. There has to be a specific type of cue that elites can offer to compel others in the political system to reconsider their outlook on issue salience. Derived from dual systems theories of emotion, the dissertation predicts that issue discussion that heightens feelings of anxiety increases the likelihood of an altered outlook on issue salience. To evaluate this prediction, time series statistical techniques are employed. The time series models evaluate whether prior change in the level of anxietybased cues by the president and the media predict changes in the level of attention the public offers to that issue. The same types of models evaluate whether this form of issue discussion by the president predicts issue dynamics of the media, and vice-versa. The several issues studied are crime, health care, poverty, and the environment. Information spanning thirty years is collected from presidential papers, general and ideological media newspaper coverage, and multiple public survey organizations. The findings suggest anxiety-based issue discussion does have the potential to guide issue attention. Prior changes in anxiety-based cues do predict future levels of attention the public provides to issues. A positive shift in anxiety cues by elites appears to have the capacity to increase public attention to issues. This increase though appears to be very small and abbreviated, suggesting limited effects. Elites do not appear to influence each other through anxiety cues.

碩士<br>國立中央大學<br>資訊工程學系<br>106<br>In recent years, deep learning plays an important role in Artificial Intelligence, which Convolutional Neural Network(CNN) has a breakthrough performance comparing with the traditional methods in image classification. Object detection is the popular issue in the image processing, and it has a lot of applications in our life, include face detection, pedestrian detection which can be used in self-driving car and the self-service store need the object detection application in product detection. There were lots of object detection research published in the world. One is SSD: Single Shot Multibox Detector, which combines predictions from multiple feature maps with different resolutions to naturally handle objects of various size. Our paper combines the advantages of two networks: multi-scale network and feature pyramid network. Proposed adding the attention mechanism to the network. This network can be trained end-to-end. In this work, based on FPNSSD network and add Attention mechanism into multi-scale network. The Attention mechanism can let the deep network learned the important area in the feature map, and gave more weight in important area. Because the attention mechanism had better performance in classification and segmentation, we add attention in the multi-scale network, hopes it have better performance in small object detection. In the experiment, FPNSSD with attention got the better performance of bonding box and classification in the small object like bird, bottle in VOC challenge 2012.

碩士<br>國立臺灣大學<br>電信工程學研究所<br>96<br>Human visual attention system is a popular topic in recent years. The human visual attention system addresses the situation of computational implementation of intentional attention in the human vision. The human visual attention system is widely applied in the design of robot or automatic intelligence. In many researches, implementations about object segmentations, object recognitions, and object detections are proposed more and more frequently. In this thesis, we mainly display two methods and implementations to simulate the human visual attention model. The output is denoted as saliency. Saliency means the place where human eyes emphasis on the most when first looking at an image. We displayed the algorithms that are widely used as the basic of the build of attention model for images. Moreover, another brand new concept of the salient model representation for videos is displayed here. Detecting moving objects in videos is an issue that people has discussed with high frequency in recent years. An algorithm for the real-time implement is now a developing and popular issue. Also, it presents a concept about the real-time moving object detection in time domain and another similar concept applied in DCT data domain in videos.

碩士<br>大葉大學<br>資訊工程學系碩士班<br>103<br>In this thesis, we proposed a video quality enhancement scheme based on visual attention model to deal with low- and high-exposure videos. The proposed scheme is composed of five parts: pre-processing, visual attention model, multilevel exposure correction, data fusion, and post-processing. To make the proposed scheme easily measure visual cues of each frame, the pre-processing procedure is used to coarsely modify each input frame. After pre-processing, visual attention model is used to extract visual features of each frame and then we conducted multi-level exposure correction for each frame according to the visual attention model. For each frame, we fuse these versions generated by multi-level exposure correction to obtain the final resulting frame. To reduce the impact of flicker on visual quality, a post-processing procedure is developed to enhance the video quality. The experiment results demonstrate that the proposed scheme can deal with videos with low and high exposures. The results also show that the proposed scheme outperforms some existing methods in terms of visual quality.

碩士<br>國立臺灣大學<br>資訊工程學研究所<br>105<br>Recommendation systems have mainly dealt with the problem of recommending items to fit user preferences, while the dynamicity of user interest is not fully considered. We observe that music streaming platforms like YouTube always recommend songs that either from the same artist or with the same title, assuming that users have a static interest in similar items, but ignore the fact that we get satiated easily with repeated consumptions. To provide a more appealing user experience, recent developments in recommendation system have focused on introducing novelty in the recommendation list; however, none of these works try to discuss ``when will the users shift their interest?", the key problem that determines our strategies to recommend new items or similar items. In this work, we present a novel model for interest shift prediction. By the state-of-the-art deep learning techniques that excel in extracting high-level knowledge, we try to construct the latent representations of mental states, and apply the attention mechanism on our model to automatically detect the shifting patterns in the listening records. Experiments and case studies show that our models can achieve good accuracy as well as interpretability.

碩士<br>國立臺灣大學<br>資訊工程學研究所<br>106<br>Creative writing has become a standard task to showcase the power of artificial intelligence. This work tackles a challenging task in this area, the lyrics rewriting. This task possesses several unique challenges. First, we require the outputs to be not only semantically correlated with the original lyrics, but also coherent in segmentation structure, rhyme as the rewritten lyrics must be performed by the artist with the same music. Second, there is no parallel rewriting lyrics corpus available for supervised training. We propose a deep neural network based model for this task and exploit both general evaluation metrics such as ROUGE and human study to evaluate the effectiveness of the model.

碩士<br>國立雲林科技大學<br>電機工程系<br>106<br>In this thesis, we develop an image caption generation based on deep learning and visual attention model. This system is composed of several parts: object detection, saliency computation, and image caption generation. In the object detection part, a deep learning technique, Faster R-CNN, is used to detect and classify objects in images. A pre-trained model can classify 80 categories for image classification. In the saliency computation, the pre-training model proposed in [8] is to compute the saliency value of each ROI image. According to category information and saliency value, the proposed system can generate the corresponding image caption. To evaluate the performance of the proposed system, the COCO 2014 image set is used. There are 30,000 images in the COCO 2014 image set. For image caption, the BLEU value of the proposed system is higher than that of [11]. Experimental results show that the proposed system is superior to the existing method [11].

碩士<br>國立中央大學<br>軟體工程研究所<br>107<br>Smart transportation is a crucial issue for a smart city, and the forecast for taxi demand is one of the important topics in smart transportation. If we can effectively predict the taxi demand in the near future, we may be able to reduce the taxi vacancy rate, reduce the waiting time of the passengers, increase the number of trip counts for a taxi, expand driver’s income, and diminish the power consumption and pollution caused by vehicle dispatches. This paper proposes an efficient taxi demand prediction model based on state-of-the-art deep learning architecture. Specifically, we use the LSTM model as the foundation, because the LSTM model is effective in predicting time-series datasets. We enhance the LSTM model by introducing the attention mechanism such that the traffic during the peak hour and the off-peak period can better be predicted. We leverage a multi-layer architecture to increase the predicting accuracy. Additionally, we design a loss function that incorporates both the absolute mean-square-error (which tends under-estimate the low taxi demand areas) and the relative meansquare-error (which tends to misestimate the high taxi demand areas). To validate our model, we conduct experiments on two real datasets — the NYC taxi demand dataset and the Taiwan Taxi’s taxi demand dataset in Taipei City. We compare the proposed model with non-machine learning based models, traditional machine learning models, and deep learning models. Experimental results show that the proposed model outperforms the baseline models.

碩士<br>國立雲林科技大學<br>電機工程系<br>106<br>In this thesis, we develop a hash code generation based on deep learning and visual attention model for image retrieval. This system is composed of several parts: object detection, saliency computation, and hash code generation. In the object detection part, a deep learning technique, Faster R-CNN, is used to detect and classify objects in images. A pre-trained model can classify 20 categories for image classification. In the saliency computation, the pre-training model proposed in [26] is to compute the saliency value of each object. According to category information and saliency value, the proposed system can generate the corresponding hash code. To evaluate the performance of the proposed system, the PASCAL VOC image set is used. There are 27088 images in the PASCAL VOC image set. For image retrieval, the nDCG value of the proposed system is higher than that of [29]. Experimental results show that the proposed system is superior to the existing method [29]. Keywords: object detection, visual attention, hash code

碩士<br>中國文化大學<br>國際貿易學系<br>104<br>Banner is the most common way to advertise on the Internet. Previous studies suggested that consumers seem to avoid looking at banners when they visit the website. For consumers, the frame of website has already become a schema in their brain and made them ignore unimportant information in their visual field. Nowadays, in the advanced networking, we need to consider that the problem: how to make them look at the banner again and increase the internet advertising effect.There are two purposes in the present study. Firstly, do participants get higher gaze behavior on the banner through the effect to arousal and attention on schema incongruity when the shape of banner changed? Secondly, due to the construction of schema was accumulating by time, experiences and learning, we are curious about the question that do the change of banner decrease consumer`s gaze behavior by increasing the exposure. In this thesis, we use the experimental design. In the experiment one, we examined eyes movements, including number and duration of fixations based on manipulating the consistency of the consumer’s schema and the shapes of the advertising banner and schema. In experiment two, we’ll make the participants who had viewed the banners that incongruity of consumer schema to review the banner that the incongruity of consumers schema again, then, examine the number and the duration of gaze behavior after experiment one. We’ll compare what is the difference between the first time while watching the banners and the last time. Through this thesis, we can affirm that it is possible to retain the explosion, increase the effect of banners by changing the consumers’ schema. However, the banners must change constantly or the effect of the advertisement still will decrease eventually.

碩士<br>國立成功大學<br>行為醫學研究所<br>97<br>Objective： This study expects to develop a simple attention-screening questionnaire (The Student Attention Rating Scale) and establish a model to screen and intervene the students with ADHD in junior high schools. The study also assesses the attention deficit patterns on those teenagers with ADHD and explores the prevalence of ADHD subtypes. Finally, the study develops a pilot training program to assist adolescent with ADHD to improve attention performance. Method：At the beginning, there were 675 7th grade students participating the primary screening by using The Student Attention Rating Scale, teachers’ reports and reviewing medical history related to attention problems. The participants were gradually given the elaborate questionnaire, neuropsychological test and interview conducted by Psychiatrists. In order to confirm the appropriate use of the Student Attention Screening Scale, the screening results have been crossly reviewed with the psychiatric reports. At the end, investigating the prevalence of different subtype of ADHD and administrating individual attention training programs to explore an available school-based intervention model of adolescent with ADHD. Results：There were 23 students being diagnosed with ADHD in this study. Six out of 23 students showed medical history because of attention problem. Seven students were reported as possible ADHD by their teachers. The rest of the students were found in ADHD by the Student Attention Rating Scale. Among these 23 students, the ratio of ADHD-I, ADHD-H and ADHD-C is 17：1：5. The estimated prevalence of ADHD in adolescence is 9.42%. The sensitivity of the screening model is 0.671, and the specificity is 0.887. The teenagers with ADHD present poor sustained attention, require long reaction time and perform insufficient attention resource. The reaction time and the correct reaction rate of the subjects were in improved after the individual attention training programs. Conclusions：According to the prevalence of this study, there are still high percentage of ADHD occurred in adolescent. But many of them have not been found. In this study, the occurrence of ADHD-I is 3.4 times greater than ADHD-C. However, the diagnosis ratio is 17% and 40%. It shows that there are fewer ADHD-I being diagnosed and treated. The screening model suggested in this study is simple, but effectively discovers 82.6% adolescent with ADHD who were ignored by parents and teachers. This study also suggested an available school-based neuropsychological attention assessment and training model for school clinical psychologists.

Les interventions proactives ou comportementales en classe sont reconnues empiriquement pour leur efficacité à améliorer le comportement ou le rendement scolaire des enfants ayant un TDAH (DuPaul & Eckert, 1997; Hoza, Kaiser, & Hurt, 2008; Pelham & Fabiano, 2008; Zentall, 2005). Or, l’écart entre les interventions probantes et celles retrouvées dans le milieu général de l’éducation souligne l’importance de répliquer les résultats d’études obtenus dans un environnement contrôlé dans un format de livraison réaliste. L’objectif principal de cette thèse est d’élaborer et d’évaluer un programme de consultation individuelle (PCI) fondé sur une démarche de résolution de problème et d’évaluation fonctionnelle, pour soutenir les enseignants du primaire dans la planification et la mise en œuvre cohérente des interventions privilégiées pour aider les enfants ayant un TDAH. D’abord, une recension des principales modalités d’intervention auprès des enfants ayant un TDAH est effectuée afin d’identifier les interventions à inclure lors du développement du programme. Par la suite, des solutions favorisant le transfert des interventions probantes à la classe ordinaire sont détaillées par la proposition du PCI ayant lieu entre un intervenant psychosocial et l’enseignant. Enfin, l’évaluation du PCI auprès de trente-sept paires enfant-enseignant est présentée. Tous les enfants ont un diagnostic de TDAH et prennent une médication (M). Les parents de certains enfants ont participé à un programme d’entraînement aux habiletés parentales (PEHP). L’échantillon final est: M (n = 4), M et PEHP (n = 11), M et PCI (n = 11), M, PEHP et PCI (n = 11). Les résultats confirment l’efficacité du PCI au-delà de M et M + PEHP pour éviter une aggravation des comportements inappropriés et améliorer le rendement scolaire des enfants ayant un TDAH. Par ailleurs, une augmentation de l’utilisation des stratégies efficaces par l’enseignant est observable lorsqu’il a à la fois participé au PCI et reçu une formation continue sur le TDAH en cours d’emploi. Les implications cliniques de l’intervention pour l’enfant ayant un TDAH et son enseignant de classe ordinaire sont discutées.<br>Classroom management interventions, such as behavior and academic strategies, are well-established interventions for improving social behavior and academic skills of children with ADHD (DuPaul & Eckert, 1997; Hoza, Kaiser, & Hurt, 2008; Pelham & Fabiano, 2008; Zentall, 2005). However, bridging the gap between research and practice raises the question of the practicality of interventions. Therefore, results from controlled studies need to be replicated in regular classrooms with a format that takes into account the practicality of the intervention. The aim of this research is to evaluate the effectiveness of a consultation-based program for teachers (CPT), using a problem-solving approach and a functional assessment to support elementary school teachers in the knowledge of the principles, design and implementation of classroom management evidence-based practices for children with ADHD. First, a review of the literature identifying the main interventions for ADHD children is presented. Then, the consultation-based program for regular class teachers involving solutions in the implementation of these evidence-based strategies in the classroom is detailed. Finally, the evaluation of the CPT implemented with thirty-seven child-teacher pairs is presented. All children were diagnosed as ADHD and received a stimulant medication treatment (M). The parents of some of these children had previously participated in a parent-training program (PTP). The final group composition is: M (n = 4); M + PTP (n = 11), M + CPT (n = 11), M + PTP + CPT (n = 11). Findings confirm the effectiveness of the CPT above and beyond M, and M + PTP to prevent the intensification of inappropriate behaviors and to improve academic performance of ADHD children. Results also indicate that teachers who participated in the CPT and had previous continuing education on ADHD showed a significant improvement of their classroom management strategies. Overall findings offer valuable information for discussing clinical implications for the psychosocial treatment of ADHD children.