Abstract:
Being able of recognizing objects despite changes in their appearance is a necessary skill for the vast majority of living beings. Visual systems, such the one of rats, can solve this task without an apparent effort and with surprising speed. In order to do that, the visual cortex must be able to build a representation of a specific visual stimulus that is invariant to identity-preserving transformations. In particular, the rat lateral extrastriate areas are thought to be organized as a hierarchy, each one representing features of a certain complexity. Several studies have shown that, along the V1-LM-LI-LL progression of rat visual areas, information about low-level features is gradually pruned in favor of information about high-level features. In this scenario, artificial intelligence techniques became extremely important. Information theory measures, dimensionality reduction techniques and unsupervised learning are used for respectively measuring the amount of information in neuronal responses, reducing the high dimensionality of neuronal representation and extracting its hidden structures. This thesis aims to expand the analysis of object representation in rat visual cortex, and its evolution across the extrastriate visual areas, especially using the dominant sets clustering (Pelillo et al.). The purpose of the latter is to extract maximally coherent groups satisfying internal homogeneity and external inhomogeneity properties by a sequential search of structures in the data adhering to this cluster notion.
