|M.Sc Student||Hazan Alon|
|Subject||Learning an Attention Model Artificial in an|
|Department||Department of Electrical Engineering||Supervisor||Professor Ron Meir|
|Full Thesis text|
Human visual perception of the world is of a large fixed image that is highly detailed and sharp. However, the retina is not isotropic in the sense of receptor density, a small central region called the fovea is very dense and exhibits high resolution whereas a peripheral region around it has much lower spatial resolution. Thus, contrary to our perception, we are only able to observe a very small region around the line of sight with high resolution. In order to perceive the visual world as humans do, there must exist a mechanism that controls the eyes and directs them to the numerous points of interest within the view, and also a mechanism that allows the human to perceive that patched information as a complete and stable view. Such mechanisms do exist in fact. Eye movements are unconscious motions following the demand of attention. The mission of moving the eye to an object or region of interest is done by saccadic eye movements, those are quick ballistic movements that direct the eye to new targets that require the high resolution of the Fovea. Once the target is found, the eyes fixate for a fraction of a second while the visual system extracts the necessary information. An artificial visual system was built based on a fully recurrent neural network set within a reinforcement learning protocol, and learned to attend to regions of interest while solving a classification task. The model is consistent with several experimentally observed phenomena, and suggests novel predictions.