Abstract
Learning in the ventral cortical stream leads to recognition categories that tend to be increasingly independent of object size and position at higher cortical levels. The anterior inferotemporal cortex (ITa) exhibits such invariance, which helps to prevent a combinatorial explosion in memory of object representations at every size and position. Zoccolan, Kouth, Poggio, & Dicarlo (2007, Journal of Neuroscience) showed that ITa cells demonstrate a tradeoff between object selectivity and position tolerance. A neural model is presented of how perceptual and attentive learned categorization processes in the visual cortex together generate robust quantitative simulations of these data using a combination of well-known cortical mechanisms. The model was tested using the same training and testing procedure as in Zoccolan et al. (2007) and the same realistic natural stimuli from the Cal Tech 101 dataset. The Zoccolan et al. (2007) data and our simulations thereof are contrary to recent models of IT (e.g. Riesenhuber & Poggio, 2000, Nature Neuroscience; Wallis & Rolls, 1997, Progress in Neurobiology), which propose that ITa cells have object selective and position tolerant response profiles. The current model clarifies how the Zoccolan et al. tradeoff found in ITa cells may be the natural result of basic mechanisms of the ventral cortical stream.
Supported in part by the National Science Foundation (SBE-0354378).