Limits...
High Accuracy Decoding of Dynamical Motion from a Large Retinal Population.

Marre O, Botella-Soler V, Simmons KD, Mora T, Tkačik G, Berry MJ - PLoS Comput. Biol. (2015)

Bottom Line: We show that the bar's position can be reconstructed from retinal activity with a precision in the hyperacuity regime using a linear decoder acting on 100+ cells.The classical view would have suggested that the firing rates of the cells form a moving hill of activity tracking the bar's position.As a result, population redundancy was high, and we could find multiple, disjoint subsets of neurons that encoded the trajectory with high precision.

View Article: PubMed Central - PubMed

Affiliation: Department of Molecular Biology and Neuroscience Institute, Princeton University, Princeton, United States of America; Institut de la Vision, INSERM UMRS 968, UPMC UM 80, CNRS UMR 7210, Paris, France.

ABSTRACT
Motion tracking is a challenge the visual system has to solve by reading out the retinal population. It is still unclear how the information from different neurons can be combined together to estimate the position of an object. Here we recorded a large population of ganglion cells in a dense patch of salamander and guinea pig retinas while displaying a bar moving diffusively. We show that the bar's position can be reconstructed from retinal activity with a precision in the hyperacuity regime using a linear decoder acting on 100+ cells. We then took advantage of this unprecedented precision to explore the spatial structure of the retina's population code. The classical view would have suggested that the firing rates of the cells form a moving hill of activity tracking the bar's position. Instead, we found that most ganglion cells in the salamander fired sparsely and idiosyncratically, so that their neural image did not track the bar. Furthermore, ganglion cell activity spanned an area much larger than predicted by their receptive fields, with cells coding for motion far in their surround. As a result, population redundancy was high, and we could find multiple, disjoint subsets of neurons that encoded the trajectory with high precision. This organization allows for diverse collections of ganglion cells to represent high-accuracy motion information in a form easily read out by downstream neural circuits.

No MeSH data available.


Related in: MedlinePlus

High-accuracy reconstruction of the bar’s trajectory.A: Schematic of the linear decoding method, here for 4 cells. A temporal filter is associated with each cell. Each time the cell spikes, its filter is added to the ongoing reconstruction at the time of the spike. The filters are optimized on part of the data to have the lowest reconstruction error and then tested on the rest of the data. B: Top: prediction of the bar’s position (black) from the activity of 123 cells in the salamander retina versus the real trajectory (red). Bottom: prediction of the bar’s position (black) from the activity of 178 cells in the guinea pig retina versus the real trajectory (red). C, D: Decoding performance plotted against the number of cells in the salamander (C) and guinea pig (D). Gray points correspond to random subsets of cells, black to the average performance. E: Histogram of the average decoding performance across all experiments using either causal (black) or acausal (white) decoding filters; results shown for the entire recorded neural population in both the salamander (left) and guinea pig (right).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4489022&req=5

pcbi.1004304.g002: High-accuracy reconstruction of the bar’s trajectory.A: Schematic of the linear decoding method, here for 4 cells. A temporal filter is associated with each cell. Each time the cell spikes, its filter is added to the ongoing reconstruction at the time of the spike. The filters are optimized on part of the data to have the lowest reconstruction error and then tested on the rest of the data. B: Top: prediction of the bar’s position (black) from the activity of 123 cells in the salamander retina versus the real trajectory (red). Bottom: prediction of the bar’s position (black) from the activity of 178 cells in the guinea pig retina versus the real trajectory (red). C, D: Decoding performance plotted against the number of cells in the salamander (C) and guinea pig (D). Gray points correspond to random subsets of cells, black to the average performance. E: Histogram of the average decoding performance across all experiments using either causal (black) or acausal (white) decoding filters; results shown for the entire recorded neural population in both the salamander (left) and guinea pig (right).

Mentions: Under these stimulus conditions, the activity of ganglion cells was sparse and strongly modulated (Fig 1C and 1D). Our goal was to decode the position of the bar at all times from this retinal activity. We used a linear decoder that took as an input the spike trains of all the ganglion cells and predicted the position of the bar as a function of time [26]. A temporal filter is associated with each cell and added to the prediction each time the cell fires an action potential (Fig 2A). The filter shapes were fitted on a training fraction of the data (2/3) to minimize the squared error between the predicted and true trajectories (see Methods for details). To test the decoder, the prediction was evaluated on the previously withheld test fraction of the data (1/3).


High Accuracy Decoding of Dynamical Motion from a Large Retinal Population.

Marre O, Botella-Soler V, Simmons KD, Mora T, Tkačik G, Berry MJ - PLoS Comput. Biol. (2015)

High-accuracy reconstruction of the bar’s trajectory.A: Schematic of the linear decoding method, here for 4 cells. A temporal filter is associated with each cell. Each time the cell spikes, its filter is added to the ongoing reconstruction at the time of the spike. The filters are optimized on part of the data to have the lowest reconstruction error and then tested on the rest of the data. B: Top: prediction of the bar’s position (black) from the activity of 123 cells in the salamander retina versus the real trajectory (red). Bottom: prediction of the bar’s position (black) from the activity of 178 cells in the guinea pig retina versus the real trajectory (red). C, D: Decoding performance plotted against the number of cells in the salamander (C) and guinea pig (D). Gray points correspond to random subsets of cells, black to the average performance. E: Histogram of the average decoding performance across all experiments using either causal (black) or acausal (white) decoding filters; results shown for the entire recorded neural population in both the salamander (left) and guinea pig (right).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4489022&req=5

pcbi.1004304.g002: High-accuracy reconstruction of the bar’s trajectory.A: Schematic of the linear decoding method, here for 4 cells. A temporal filter is associated with each cell. Each time the cell spikes, its filter is added to the ongoing reconstruction at the time of the spike. The filters are optimized on part of the data to have the lowest reconstruction error and then tested on the rest of the data. B: Top: prediction of the bar’s position (black) from the activity of 123 cells in the salamander retina versus the real trajectory (red). Bottom: prediction of the bar’s position (black) from the activity of 178 cells in the guinea pig retina versus the real trajectory (red). C, D: Decoding performance plotted against the number of cells in the salamander (C) and guinea pig (D). Gray points correspond to random subsets of cells, black to the average performance. E: Histogram of the average decoding performance across all experiments using either causal (black) or acausal (white) decoding filters; results shown for the entire recorded neural population in both the salamander (left) and guinea pig (right).
Mentions: Under these stimulus conditions, the activity of ganglion cells was sparse and strongly modulated (Fig 1C and 1D). Our goal was to decode the position of the bar at all times from this retinal activity. We used a linear decoder that took as an input the spike trains of all the ganglion cells and predicted the position of the bar as a function of time [26]. A temporal filter is associated with each cell and added to the prediction each time the cell fires an action potential (Fig 2A). The filter shapes were fitted on a training fraction of the data (2/3) to minimize the squared error between the predicted and true trajectories (see Methods for details). To test the decoder, the prediction was evaluated on the previously withheld test fraction of the data (1/3).

Bottom Line: We show that the bar's position can be reconstructed from retinal activity with a precision in the hyperacuity regime using a linear decoder acting on 100+ cells.The classical view would have suggested that the firing rates of the cells form a moving hill of activity tracking the bar's position.As a result, population redundancy was high, and we could find multiple, disjoint subsets of neurons that encoded the trajectory with high precision.

View Article: PubMed Central - PubMed

Affiliation: Department of Molecular Biology and Neuroscience Institute, Princeton University, Princeton, United States of America; Institut de la Vision, INSERM UMRS 968, UPMC UM 80, CNRS UMR 7210, Paris, France.

ABSTRACT
Motion tracking is a challenge the visual system has to solve by reading out the retinal population. It is still unclear how the information from different neurons can be combined together to estimate the position of an object. Here we recorded a large population of ganglion cells in a dense patch of salamander and guinea pig retinas while displaying a bar moving diffusively. We show that the bar's position can be reconstructed from retinal activity with a precision in the hyperacuity regime using a linear decoder acting on 100+ cells. We then took advantage of this unprecedented precision to explore the spatial structure of the retina's population code. The classical view would have suggested that the firing rates of the cells form a moving hill of activity tracking the bar's position. Instead, we found that most ganglion cells in the salamander fired sparsely and idiosyncratically, so that their neural image did not track the bar. Furthermore, ganglion cell activity spanned an area much larger than predicted by their receptive fields, with cells coding for motion far in their surround. As a result, population redundancy was high, and we could find multiple, disjoint subsets of neurons that encoded the trajectory with high precision. This organization allows for diverse collections of ganglion cells to represent high-accuracy motion information in a form easily read out by downstream neural circuits.

No MeSH data available.


Related in: MedlinePlus