Hierarchical novelty-familiarity representation in the visual cortex

Abstract

Event Abstract Back to Event Hierarchical novelty-familiarity representation in the visual cortex Olshausen and Field (1996, 1997) and Rao and Ballard (1999) have shown that the statistics of natural images can explain some receptive field (RF) properties of simple cells in V1. However, only small patches of images were used, with many cells reading out multiple statistics from each RF. Furthermore, no attempt was made to evaluate how well the read-outs thus obtained encoded the stimuli. Predictive coding used by Rao and Ballard is an instance of a wider class of generative Bayesian models that has been suggested to serve as an organizing principle for the entire cortical hierarchy (Lee and Mumford, 2003; Friston, 2005). Thus, we investigate here how good predictive coding actually is at coding whole images using a natural topographic connectivity, where no two cells have exactly the same RF, but adjacent cells' RFs have strong overlaps. In predictive coding, feedback from higher areas carries expectations of lower-level activity (familiarity signal), whereas the feedforward (novelty) signals carry discrepancies between the expectations and the stimuli. Visual recognition becomes an iterative process relaxing to a solution that matches experience with sensory input. We use two interconnected populations of neurons, one coding for the familiarity, and the other, for novelty. These form one of multiple levels in a hierarchical processing structure. In contrast to Rao and Ballard, in our model top-down effects are confined to a single level. We suggest that this limited feedback is preferable for biologically compatible fast recognition. We use 1000 natural images (Van Hateren and van der Schaaf, 1998) for unsupervised learning of the synaptic efficacies of two visual processing levels. The coding performance is then evaluated on a set of 200 different images from the same database by reconstructing the stimulus based on the internal code in our generative model and directly comparing the prediction to the actual image. Despite a compression factor of 4 for each level, the image reconstruction quality on the test images is quite good and strongly exceeds that of local averaging, implying that the learning results in the extraction of features characteristic of the set of natural images as a whole. With our model, we have found that the extra-classical RF effect of endstopping can arise due to the topographic connectivity from interactions within the first processing level without the need for feedback from the next level, as in Rao and Ballard. Furthermore, the effective RF's after learning resemble those of simple cells in V1 and learning with the topographic map leads to a global organization of the RF's. Finally, the proposed architecture allows for the simultaneous, but separate, representation of familiarity and novelty in the visual cortex. The novelty signal could produce read-out to higher processing centers, activating them if a localized novel signal, such as a predator in a serene environment, suddenly appears in a familiar image. Thus, our implementation of predictive coding could be an effective way for the visual system to combine fast hierarchical visual representation with the interaction with higher information-processing areas in the brain. Conference: Computational and systems neuroscience 2009, Salt Lake City, UT, United States, 26 Feb - 3 Mar, 2009. Presentation Type: Poster Presentation Topic: Poster Presentations Citation: (2009). Hierarchical novelty-familiarity representation in the visual cortex. Front. Syst. Neurosci. Conference Abstract: Computational and systems neuroscience 2009. doi: 10.3389/conf.neuro.06.2009.03.096 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 02 Feb 2009; Published Online: 02 Feb 2009. Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Google Google Scholar PubMed Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.