ATTENTION MODELS FOR DEEP LEARNING

2017 edition

Carola Figueroa

In the past few years, Deep Neural Network (DNNs) achieved great success in Computational Vision Systems, such as object recognition,extracting distinctive
features from images, etc. In regard to object recognition systems, such as “bag-of-words” and “deep belief networks”, have obtained excellent results on large data sets. In parallel, human vision researchers have developed accurate models of the first layers of the human visual system.
These computational vision systems, both in object recognition as in human vision research, are common in the fact that they model the same vision processes as a purely bottom-up process. Therefore, the vast majority of existing computational vision systems are “bottom-up” and lack “top-down” feedback. Given the importance in resolved this problem, appears the idea this proposal of thesis, that intends to implement and develop a new method for recognition and object categories