A Neural Architecture for the Representation of Scenes
Christoph von der Malsburg · 1992
Abstract Learning and memory deal with the process by which knowledge and skills are absorbed and stored in the brain. To understand them it is important to have precise ideas about the stuff that is being stored: What is its nature? How is it organized? How is it implemented in neural hardware? The neuroscience of learning and memory is at present conditioned by a set of concepts based on specific answers to these questions. In this chapter I will raise a number of issues that are not solved in this framework. These issues have come up during attempts at model reconstructions of elementary capabilities of the brain, such as pattern recognition, perceptual segmentation, and scene representation. Back-engineering is an important window to the brain. It is true that if a given functional problem had a million technical solutions, the construction of any one of them would say nothing about nature’ s solution. However, there is good reason to believe that relevant functional problems have very few solutions, and any one of them will suggest important constraints to the neuroscience of brain function. Backengineering is a fruitful exercise because it necessitates whole-system solutions. This puts it in contrast to many studies that limit their view to small subtasks and subsystems, thus leaving out essential aspects.