Sparse representation in the construction of curvature selectivity in V4 Y Hatori, T Mashita, K Sakai |
|---|
Department of Computer Science, University of Tsukuba, Japan
|
| Physiological studies have reported that V4 neurons are selective to curvature and its direction, and their population preference is biased toward acute curvature [e.g., Carlson et al., 2011, Current Biology, 21, 288-293]. Although these characteristics appear crucial for the primitive representation of shape, what principle underlies such complex selectivity has not been clarified. We propose that sparse representation is crucial for the construction of the selectivity, as similar to V1 [Olshausen and Field, 1996, Nature, 381, 288-293]. To test the proposal, we applied component analysis with sparseness constraint to activities of model neurons, and investigated the dependence of basis functions on sparseness. The computed bases represent the receptive field that is generated given the constraint. The structures of the bases were localized and appeared to represent curvature when sparseness is medium to large (>0.6). To investigate whether these bases reproduce the characteristics of V4 neurons, we computed selectivity of each basis in curvature/direction domain, and their population preference, in the way same as the physiological experiments. The selectivity of bases and their population preference agreed with the physiology when sparseness was medium (0.6-0.8). These results indicate that medium-to-large sparseness is crucial for the construction of curvature selectivity in V4. |
| Up Home |
|---|