We consider the problem of detecting a large number of different classes of objects in cluttered scenes. Traditional approaches require applying a battery of different classifiers to the image, at multiple locations and scales. This can be slow and can require a lot of training data, since each classifier requires the computation of many different image features. In particular, for independently trained detectors, the (run-time) computational complexity, and the (training-time) sample complexity, scales linearly with the number of classes to be detected. It seems unlikely that such an approach will scale up to allow recognition of hundreds or thousands of objects. We present a multi-class boosting procedure (joint boosting) that reduces the computational and sample complexity, by finding common features that can be shared across the classes (and/or views). The detectors for each class are trained jointly, rather than independently. For a given performance level, the total number of features required, and therefore the computational cost, is observed to scale approximately logarithmically with the number of classes. The features selected jointly are closer to edges and generic features typical of many natural structures instead of finding specific object parts. Those generic features generalize better and reduce considerably the computational cost of an algorithm for multi-class object detection.

We consider the problem of detecting a large number of different classes of objects in cluttered scenes. Traditional approaches require applying a battery of different classifiers to the image, at multiple locations and scales. This can be slow and can require a lot of training data, since each classifier requires the computation of many different image features. In particular, for independently trained detectors, the (run-time) computational complexity, and the (training-time) sample complexity, scales linearly with the number of classes to be detected. It seems unlikely that such an approach will scale up to allow recognition of hundreds or thousands of objects.We present a multi-class boosting procedure (joint boosting) that reduces the computational and sample complexity, by finding common features that can be shared across the classes (and/or views). The detectors for each class are trained jointly, rather than independently. For a given performance level, the total number of features required, and therefore the computational cost, is observed to scale approximately logarithmically with the number of classes. The features selected jointly are closer to edges and generic features typical of many natural structures instead of finding specific object parts. Those generic features generalize better and reduce considerably the computational cost of an algorithm for multi-class object detection.

Knowledge sharing across domains is key to bringing down the cost of production and the time to market of products. This thesis is directed to improve the knowledge sharing capability of the present systems that use information and communication technologies. Systems for different domains have structures that are made up of concepts and relations with different semantic interpretations. Therefore, knowledge sharing across such domains becomes an issue. Knowledge sharing across multiple domains can be facilitated through a system that can provide a shared understanding across multiple domains. This requires a rigorous common semantic base underlying the domains across which to share knowledge.

This research is embedded in the Chomskyan linguistic theory, specifically in the Principles and Parameters framework (Chomsky, 1981 and following) and its more recent reformulations, namely, the Minimalist Program (Chomsky, 1995 and following) and it has as main objectives to analyze and to explain the acquisition process of grammatical number agreement between the elements of the DP in Brazilian Portuguese and to explain the diversity found in the speech data of children with regard to the realization of grammatical number. The acquisition data presented in this investigation are the spontaneous production of two children in the acquisition phase of the city of Maceió-AL, one male and one female, with 2, 11 and 3, 5 respectively. Data from other languages are derived from other studies have made. We hypothesize that the information corresponding to the number on the items that form the functional category Determinant (D) is of fundamental importance for the identification number of the grammatical system of the Portuguese, being the decisive factor responsible for the core agreement between the nominal number of elements of DP in PB. Put another way, the core where the number would be interpretable trace is a parameter to be fixed and the core in determining the number that would be interpretable in BP. This would explain the fact that where agreement is non-redundant, BW brand morphological number appears in the determinant, the other way, ie, one in which the morpheme is adjoined only to the name, an ungrammatical form, at least in PB. This would explain the fact that where agreement is non-redundant, BW brand morphological number appears in the determinant, the other way, i. e, one in which the morpheme is adjoined only to the name, an ungrammatical form, at least in PB. Regarding the agreement between the DP flexionáveis elements, we assume as in Chomsky (1988, 1989) that this is due to a syntactic operation, Agree, but recast in Frampton & Gutmann (2000a). With this reformulation, it is assumed as a sharing agreement between core syntactic features. The results of the analysis of language production of children acquiring the CP, we can conclude that the appearance of determinants can be defined as a trigger for the parameter of agreement nominal internal elements of the DP, the trace of definiteness is crucial for setting the parameter agreement, and the co-occurrence in the speech patterns of children from non-redundant and redundant grammatical and even ungrammatical pattern. We conclude that this fact can be explained due to the variation of the input to which the child is exposed, ie, input variable has also resulted in an output variable. Thinking of PB in the data, we conclude that D is, being the head of the chain-б responsible for definiteness and semantic interpretation at LF, bearing the mark of plural. Where have varieties of brand PB morphology on all elements of the phrase, we can consider that the brand is spreading because of the sharing features. / Fundação de Amparo a Pesquisa do Estado de Alagoas / Esta pesquisa está inserida na Teoria Linguística Chomskyana, mais especificamente no modelo de Princípios e Parâmetros (Chomsky, 1981 e seguintes) e em suas reformulações mais recentes, a saber, o Programa Minimalista (Chomsky, 1995 e seguintes) e apresenta como principais objetivos analisar e explicar o processo de aquisição da concordância de número gramatical entre os elementos do DP no Português Brasileiro e explicar a diversidade encontrada nos dados de fala de crianças no que diz respeito à realização gramatical de número. Os dados de aquisição apresentados nesse trabalho são de produção espontânea de duas crianças em fase de aquisição da cidade de Maceió- AL, uma do sexo masculino e outra do sexo feminino, com 2;11 e 3;5 respectivamente. Os dados de outras línguas são retirados de outros trabalhos já constituídos. Hipotetizamos que a informação correspondente a número constante nos itens que formam a categoria funcional Determinante (D) é de fundamental importância para a identificação do sistema gramatical de número no português, sendo o Determinante o núcleo responsável pela concordância nominal de número entre os elementos do DP no PB. Dizendo de outro modo, o núcleo em que o traço de número seria interpretável é um parâmetro a ser fixado e o determinante o núcleo em que o número seria interpretável no PB. Isso explicaria o fato de que nos casos em que ocorre concordância não-redundante em PB, a marca morfológica de número aparecer no determinante, sendo a outra forma, ou seja, aquela em que o morfema é adjungido apenas ao nome, uma forma agramatical, pelo menos em PB. No que tange à concordância entre os elementos flexionáveis do DP, assumimos assim como em Chomsky (1988, 1989) que essa acontece devido a uma operação sintática, Agree, mas reformulada por Frampton & Gutmann (2000a). Com essa reformulação, assume-se concordância como compartilhamento de traços entre núcleos sintáticos. Com os resultados das análises das produções linguísticas das crianças adquirindo o PB, podemos concluir que o aparecimento de determinantes definidos possa ser um dado acionador para o parâmetro da concordância nominal interna aos elementos do DP, sendo o traço de definitude crucial para a fixação do parâmetro de concordância, além da co-ocorrência na fala das crianças dos padrões gramaticais redundantes e não redundantes e ainda do padrão agramatical. Concluímos que esse fato é explicável devido à variação do input ao qual a criança está exposta, ou seja, input variável tem como conseqüência um output também variável. Pensando ainda nos dados do PB, concluímos que é D, por ser a cabeça da cadeia- б o responsável pela definitude e pela interpretação semântica em LF, carregando a marca de plural. Nos casos em que as variedades do PB apresentam marca morfológica em todos os elementos do sintagma, podemos considerar que a marca se alastra por conta do compartilhamento de traços.