Learning W-operators in the boolean interval partition lattice learning space

• Ano de defesa: 2024
• Autor(a) principal: Cunha Filho, Francisco Edvar da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Algoritmo ISI; Busca estocástica; Descida Estocástica no Reticulado Booleano das Partições Intervalares para Aprendizado (SDBIPL); ISI algorithm; Problema da curva U; Stochastic Descent on the Boolean Interval Partition Lattice (SDBIPL); Stochastic search; U-curve problem; W-operadores; W-operators
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/45/45134/tde-11122024-212858/

This work presents the Stochastic Descent on the Boolean Interval Partition Lattice (SDBIPL) algorithm, a novel contribution to the field of lattice-based learning. The SDBIPL algorithm extends the U-curve problem to the Boolean Interval Partition Lattice (BIPL), offering a more flexible and adaptive search strategy compared to the pioneering ISI algorithm. By conducting a stochastic search and dynamically adjusting the complexity of the learned function, SDBIPL overcomes the limitations of deterministic exploration and addresses the challenge of uncontrolled complexity inherent in ISI. We provide an efficient C++ implementation of the SDBIPL algorithm and demonstrate its effectiveness in learning W-operators for image noise removal. The promising results highlight the potential of the proposed approach. Future work will explore applications of the SDBIPL algorithm to more complex problems involving W-Operators and classification, such as digit recognition. Furthermore, we aim to optimize the algorithm through parallelization and refined error calculation techniques to enhance its performance and scalability.