Human Processes in Artificial Vision

Deep learning has exhibited unparalleled success over recent years. Simple building blocks inspired by the human brain that have been in development since the 1940s have, for the first time, outperformed traditional machine learning methods. However, the descent into the deep has not been without criticism. Some believe that these models are too similar to the brain, making them inefficient and difficult to interpret. Others believe that models are not similar enough and lack the complexity required to truly reflect our most incomprehensible biological system. We take the view that while these systems have been inspired by the actions of neurons at a cellular level, they have often failed to capture some of the high level processes evident in human intelligence. We build upon previous research in the area and use observations from neuroscience and psychology to augment artificial models of vision with functionality previously held in the exclusive purview of humanity.