A dynamic model for decoding direction and orientation in macaque primary visual cortex
- Wahiba Taouali, Giacomo Benvenuti, Frédéric Chavane, Laurent Perrinet. A dynamic model for decoding direction and orientation in macaque primary visual cortex, URL . In Proceedings of VSS, 2015 abstractUsing observations of spiking activity in a population of neurons from macaque primary visual area, we studied simultaneously the dynamics of direction and orientation decoding. Stimuli consisted of oriented bars moving in 12 different directions, the orientation being orthogonal to the direction. Bars move from 3 degrees before to 1.5 degrees after the receptive field center (RFc). We first fitted the neuronal responses to the different directions using a modified von Mises tuning function. Then, we extended this fit with respect to time in order to study the dynamics of the orientation and direction tuning. The whole temporal dynamics could be described by a simple temporal tuning model based on the combination of an unselective evoked gain, direction and orientation tuning functions and a temporal Gaussian activation profile. To evaluate this model, we adapted a maximum likelihood decoder as proposed in [Jazayeri et al. Nature Neuroscience 2006] with a leave-one-out cross validation paradigm that allows to decode the direction and the normalized population activation state. The results revealed that at the beginning of the population activation, orientation is significantly decoded but not direction (i.e. the estimated direction corresponds either to the stimulus direction or its opposite). Then, when the activation is maximal, direction is significantly decoded (area under ROC curve >0.7) despite the low values of cells' direction index (only 3% with DI >0.7). Finally, when the activation decreases (i.e. the stimulus leaves the RF), direction selectivity faded first. Our temporal tuning model reveals that the decoding is highly dynamic and evolves with the global population activation state. This is at odds to static decoding commonly used for early visual areas. Our results also show that direction can be decoded accurately from a poorly selective neuronal population..
All material (c) L. Perrinet. Please check the copyright notice.
This work was supported by European Union project Number FP7-269921, "BrainScales".