Fitting techniques to knowledge discovery through stochastic models

• Ano de defesa: 2016
• Autor(a) principal: Assunção, Joaquim Vinicius Carvalho
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Faculdade de Informática; Brasil; PUCRS; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: SIMULAÇÃO E MODELAGEM EM COMPUTADORES; REDES DE AUTÔMATOS ESTOCÁSTICOS; PROCESSOS ESTOCÁSTICOS; CADEIAS DE MARKOV - COMPUTAÇÃO; INFORMÁTICA; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/7179

Stochastic models might be useful for creating compact representations of non-deterministic scenarios. Furthermore, simulations applied to a compact model, are faster and require fewer computational resources than the use of data mining techniques over large volumes of data. The challenge is to build such models. The accuracy as well as the time and the amount of resources used to fit such models, are the key factors related to their utility. We use machine learning techniques for the fitting of structures characterized by a Markov property; especially, complex formalisms such as Hidden Markov Models (HMM) and Stochastic Automata Networks (SAN). Regarding the accuracy, we considered the state of the art on fitting techniques and model measurements based on likelihood. Regarding the computational resources, we used time series and dimensionality reduction techniques to avoid the space state explosion. Such techniques are demonstrated in a process that embodies a set of common steps for the model fitting through time series. Similar to the knowledge discovery in databases (KDD), yet using stochastic models as a main component.