ICT - Outras produções
URI Permanente para esta coleção
Navegar
Submissões Recentes
- ItemO método de Levenberg-Marquadt estocástico aplicado ao treinamento de redes neurais artificiais(2020-07-30) Benatti, Kléber; Bueno, Luis Felipe Cesar da Rocha [UNIFESP]; Nazaré, TiagoEste trabalho apresenta resultados referentes ao TCC do primeiro autor no curso de especialização em Data Science financiado pelo Itaú-Unibanco. O método de Levenberg-Marquadt tem mostrado bons resultados na resolução de problemas de quadrados mínimos não linear, pois alia a convergência do método de Newton utilizando apenas informação de primeira ordem e a boa definição de todos os seus iterandos. Sendo assim, uma aplicação natural desta técnica seria utilizá-la para minimização da função custo associado ao treinamento de redes neurais artificiais. Porém, o cálculo da matriz Jacobiana associada ao sistema pode ser muito caro quando o número de instâncias é muito alto, o que torna a otimização muito lenta. Desta forma, neste trabalho é proposto um método do tipo Levenberg-Marquadt estocástico para a minimização de funções custo associadas às redes neurais. O desempenho do algoritmo é comparado com o método de Levenberg-Marquadt clássico, além do método Adam, que é usualmente aplicado neste contexto.
- ItemEffect of chronic stress and aerobic exercise on glycemic control in rats(Federation Amer Soc Exp Biol, 2016) Marcondes, Fernanda Klein; Sanches, Andrea; Costa, Rafaela; Casarini, Dulce Elena [UNIFESP]; Cunha, Tatiana Sousa [UNIFESP]
- ItemProduction of rotary jet spun ultrathin fibers of poly-butylene adipate-co- terephthalate (PBAT) filled with nanocomposites(Spie-Int Soc Optical Engineering, 2017) Andrade, P. O. [UNIFESP]; Santo, A. M. E. [UNIFESP]; Costa, M. M.; Lobo, A. O.Composite fibers of bioabsorbable poly-butylene adipate-co-terephthalate (PBAT) reinforced with superhydrophilic carbon nanotubes and hydroxyapatite nanocrystals were obtained by rotary jet spinning technique (RJS). The fibers were morphologically and biologically analyzed and found of potential use as scaffold for hard tissue engineering.
- ItemStructural behavior of coal obtained from Kraft lignin at different carbonizing rates(Elsevier Science Bv, 2017) Brazil, Tayra R. [UNIFESP]; Baldan, Mauricio R.; Massi, Marcos [UNIFESP]; Rezende, Mirabel C. [UNIFESP]Biomass is a renewable resource which importance has attracted much attention face to the concerns involving the environmental and the oil crisis. The aim of this work is to convert lignin into coal by carbonization heat treatment of this biomass up to 1000 degrees C, under inert atmosphere, at different heating rates. The resulted coal samples were characterized by X-Ray diffraction, Raman spectroscopy and surface area analyzes. Raman spectroscopy and X-ray diffraction results show that longer heat treatment resulted in coals more ordered structurally. On the order hand, longer heat treatments favored the coal production with smaller surface area. (C) 2016 Elsevier Ltd. All rights reserved.
- ItemEdge Detection Robust to Intensity Inhomogeneity: A 7T MRI Case Study(Springer International Publishing Ag, 2017) Cappabianco, Fabio A. M. [UNIFESP]; Lellis, Lucas Santana [UNIFESP]; Miranda, Paulo; Ide, Jaime S.; Mujica-Parodi, Lilianne R.Edge detection is a fundamental operation for computer vision and image processing applications. As of 1986, John Canny proposed a methodology that became known due to its simplicity, small number of parameters, and high accuracy. The method was designed to optimally detect, locate, and trace single edges over each local gradient maximum. Since then, a number of works were proposed but none of these improvements were capable of dealing with non-uniform intensity, which are notably present in ultra high field magnetic resonance imaging (MRI). In this paper, we evaluate the effects of inhomogeneity correction over automatic edge detection methods over 7T MRI. Importantly, we propose a non-supervised edge detection method which improves the accuracy of state of the art in 28.0% as detecting head and brain edges.