Banca de DEFESA: JOÃO CARLOS PANDOLFI SANTANA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : JOÃO CARLOS PANDOLFI SANTANA
DATA : 02/09/2021
HORA: 14:00
LOCAL: virtual
TÍTULO:

Transfer Learning with Retinographic Image

PÁGINAS: 50
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
SUBÁREA: Metodologia e Técnicas da Computação
ESPECIALIDADE: Processamento Gráfico (Graphics)
RESUMO:

In recent years, some Deep Artificial Neural Network models have reached a level of accuracy comparable to medical imaging specialists in classification problems. One remaining challenge in training this kind of networks in medical image classification is the limited size of the typical training set. Therefore, training a network for medical image classification usually involves a transfer step, in which learned parameters are reused from a primary network trained in some other domain. To date, few studies have been published that investigate the properties of transfer learning in this context. The most common medical imaging modalities differ significantly from the natural images used to train the primary network, with respect to resolution and attributes such as contrast, texture, noise, etc. This suggests that parameter reuse may not be the best way to speed up network training or may even hinder its final performance. To investigate the properties of transfer learning, we propose a series of experiments in the classification of retinal fundus images, in the context of glaucoma detection and using different transfer learning configurations.

MEMBROS DA BANCA:
Presidente - Interno ao Programa - 1600876 - FRANCISCO DE ASSIS ZAMPIROLLI
Membro Titular - Examinador(a) Externo ao Programa - 2357820 - ANDERSON GABRIEL SANTIAGO CRAVO
Membro Titular - Examinador(a) Externo ao Programa - 2418478 - JOHN ANDREW SIMS
Membro Suplente - Examinador(a) Externo ao Programa - 1948426 - ALEXANDRE NOMA
Membro Suplente - Examinador(a) Externo ao Programa - 1849928 - CARLOS DA SILVA DOS SANTOS