Machine learning algorithms have shown their abilities to tackle difficult recognition problems, but they are still rife with challenges. Among these challenges is how to deal with problems where new categories constantly occur, and the datasets can dynamically grow. Most contemporary learning algorithms developed to this point are governed by the assumptions that all testing data classes must be the same as training data classes, often with equal distribution. Under these assumptions, machine-learning algorithms can perform very well, using their ability to handle large feature spaces and classify outliers. The systems under these assumptions are called Closed Set Recognition systems (CSR). However, these assumptions cannot reflect practical applications in which out-of-set data may be encountered. This adversely affects the recognition prediction performances. When samples from a new class occur, they will be classified as one of the known classes. Even if this sample is far from any of the training samples, the algorithm may classify it with a high probability, that is, the algorithm will not only be wrong, but it may also be very confident in its results. A more practical problem is Open Set Recognition (OSR), where samples of classes not seen during training may show up at testing time. Inherently, there is a problem how the system can identify the novel sound classes and how the system can update its models with new classes. This thesis highlights the problems of multi-class recognition for OSR of sounds as well as incremental model adaptation and proposes solutions towards addressing these problems. The proposed solutions are validated through extensive experiments and are shown to provide improved performance over a wide range of openness values for sound classification scenarios.

Orientadores: Anderson de Rezende Rocha, Ricardo da Silva Torres / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-25T19:52:29Z (GMT). No. of bitstreams: 1 MendesJunior_PedroRibeiro_M.pdf: 10648148 bytes, checksum: 314a33c9bb6fb8a188bfa762899107e8 (MD5) Previous issue date: 2014 / Resumo: Em reconhecimento de padrões, um cenário aberto é aquele em que não há amostras de treinamento para algumas classes que podem aparecer durante o teste. Normalmente, muitas aplicações são inerentemente de cenário aberto. Consequentemente, as soluções bem sucedidas da literatura para cenário fechado nem sempre são adequadas para problemas de reconhecimento na prática. Nesse trabalho, propomos um novo classificador multiclasse para cenário aberto, que estende o classificador Optimum-Path Forest (OPF). O OPF é um classificador de padrões baseado em grafos, simples, independente de parâmetros, multiclasse e desenvolvido para para problemas de cenário fechado. O método que propomos, o Open-Set OPF (OSOPF), incorpora a capacidade de reconhecer as amostras pertencentes às classes que são desconhecidas no tempo de treinamento, sendo adequado para reconhecimento em cenário aberto. Além disso, propomos novas medidas para avaliação de classificadores propostos para problemas em cenário aberto. Nos experimentos, consideramos seis grandes bases de dados com diferentes cenários de reconhecimento e demonstramos que o OSOPF proposto supera significativamente as abordagens existentes na literatura / Abstract: An open-set recognition scenario is the one in which there are no a priori training samples for some classes that might appear during testing. Usually, many applications are inherently open set. Consequently, the successful closed-set solutions in the literature are not always suitable for real-world recognition problems. Here, we propose a novel multiclass open-set classifier that extends upon the Optimum-Path Forest (OPF) classifier. OPF is a graph-based, simple, parameter independent, multiclass, and widely used classifier for closed-set problems. Our proposed Open-Set OPF (OSOPF) method incorporates the ability to recognize samples belonging to classes that are unknown at training time, being suitable for open-set recognition. In addition, we propose new evaluation measures for assessing the effectiveness performance of classifiers in open-set problems. In experiments, we consider six large datasets with different open-set recognition scenarios and demonstrate that the proposed OSOPF significantly outperforms its counterparts of the literature / Mestrado / Ciência da Computação / Mestre em Ciência da Computação

No description available.

No description available.