Limits...
Visual nonclassical receptive field effects emerge from sparse coding in a dynamical system.

Zhu M, Rozell CJ - PLoS Comput. Biol. (2013)

Bottom Line: Models seeking to explain these non-classical receptive field (nCRF) effects in terms of circuit mechanisms, input-output descriptions, or individual visual tasks provide limited insight into the functional significance of these response properties, because they do not connect the full range of nCRF effects to optimal sensory coding strategies.Furthermore, when the population diversity of an nCRF effect has also been reported in the literature, we show that this model produces many of the same population characteristics.These results show that the sparse coding hypothesis, when coupled with a biophysically plausible implementation, can provide a unified high-level functional interpretation to many response properties that have generally been viewed through distinct mechanistic or phenomenological models.

View Article: PubMed Central - PubMed

Affiliation: Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States of America.

ABSTRACT
Extensive electrophysiology studies have shown that many V1 simple cells have nonlinear response properties to stimuli within their classical receptive field (CRF) and receive contextual influence from stimuli outside the CRF modulating the cell's response. Models seeking to explain these non-classical receptive field (nCRF) effects in terms of circuit mechanisms, input-output descriptions, or individual visual tasks provide limited insight into the functional significance of these response properties, because they do not connect the full range of nCRF effects to optimal sensory coding strategies. The (population) sparse coding hypothesis conjectures an optimal sensory coding approach where a neural population uses as few active units as possible to represent a stimulus. We demonstrate that a wide variety of nCRF effects are emergent properties of a single sparse coding model implemented in a neurally plausible network structure (requiring no parameter tuning to produce different effects). Specifically, we replicate a wide variety of nCRF electrophysiology experiments (e.g., end-stopping, surround suppression, contrast invariance of orientation tuning, cross-orientation suppression, etc.) on a dynamical system implementing sparse coding, showing that this model produces individual units that reproduce the canonical nCRF effects. Furthermore, when the population diversity of an nCRF effect has also been reported in the literature, we show that this model produces many of the same population characteristics. These results show that the sparse coding hypothesis, when coupled with a biophysically plausible implementation, can provide a unified high-level functional interpretation to many response properties that have generally been viewed through distinct mechanistic or phenomenological models.

Show MeSH

Related in: MedlinePlus

Decomposition of the recurrent inputs contributing to the end-stopping effect.(A) Overall decomposition of the response into recurrent excitatory, inhibitory, and feedforward components; (B) Locations and orientations of the CRFs of cells contributing to the recurrent excitatory and inhibitory signals at different bar lengths. Only CRFs with significant influences are displayed (i.e.,  at steady state). The warmer color (yellow) represents the location and orientation of the CRFs for cells contributing to recurrent excitation, the cooler color (blue and cyan) represents the CRFs for cells contributing to recurrent inhibition. Higher contrast in the color indicates a stronger excitatory or inhibitory effect on the target cell. The black bar represents the target cell CRF. Note that as the bar length increases, the suppressive effect is mostly due to recurrent inhibition from cells that are a better description of the new stimulus (and therefore would be a more efficient stimulus description according to the sparse coding model).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3757072&req=5

pcbi-1003191-g009: Decomposition of the recurrent inputs contributing to the end-stopping effect.(A) Overall decomposition of the response into recurrent excitatory, inhibitory, and feedforward components; (B) Locations and orientations of the CRFs of cells contributing to the recurrent excitatory and inhibitory signals at different bar lengths. Only CRFs with significant influences are displayed (i.e., at steady state). The warmer color (yellow) represents the location and orientation of the CRFs for cells contributing to recurrent excitation, the cooler color (blue and cyan) represents the CRFs for cells contributing to recurrent inhibition. Higher contrast in the color indicates a stronger excitatory or inhibitory effect on the target cell. The black bar represents the target cell CRF. Note that as the bar length increases, the suppressive effect is mostly due to recurrent inhibition from cells that are a better description of the new stimulus (and therefore would be a more efficient stimulus description according to the sparse coding model).

Mentions: The recurrent interactions between cells in the sparse coding model implement a rich nonlinear response where cells compete to represent stimulus features. While it has been noted that stimuli in the CRF surround can produce sparse responses [5], [6], [8], the surprising finding of this work is that the particular form of inhibition and excitation necessary to implement a sparse coding model is sufficient to explain so many individual and population nCRF properties. At a high level, these effects likely arise from the present model because the observed responses produce a more efficient representation of the stimulus than alternative population responses. While a detailed investigation of how the network interactions give rise to the response properties is an interesting open question for future investigation, in general this is difficult to determine due to the interactions between the network dynamics and the stimulus dynamics (i.e., the response properties arise from the average response over a drifting grating, in addition to being influenced by network dynamics). In the case of end-stopping, the stimuli is not drifting and we can see more explicitly how this effect arises from the principles of sparse coding. In response to a given fixed stimuli, the steady-state network response is composed of a combination of feedforward excitation, recurrent excitation and recurrent inhibition. When plotting these three components of the steady-state response as a function of the bar length (Fig. 9A), it is evident that the overall response is mostly driven by the feedforward component and the recurrent inhibition. The feedforward excitation saturates as a result of the stimulus growing out of the CRF, but the recurrent inhibition keeps growing with increased bar length. To see the spatial extent of the recurrent influence, Fig. 9B shows the CRF locations and orientations of the cells influencing the target cell. As expected, inhibition mostly comes from cells with overlapping and co-linear CRFs that represent a more efficient description of the stimulus as the bar length increases.


Visual nonclassical receptive field effects emerge from sparse coding in a dynamical system.

Zhu M, Rozell CJ - PLoS Comput. Biol. (2013)

Decomposition of the recurrent inputs contributing to the end-stopping effect.(A) Overall decomposition of the response into recurrent excitatory, inhibitory, and feedforward components; (B) Locations and orientations of the CRFs of cells contributing to the recurrent excitatory and inhibitory signals at different bar lengths. Only CRFs with significant influences are displayed (i.e.,  at steady state). The warmer color (yellow) represents the location and orientation of the CRFs for cells contributing to recurrent excitation, the cooler color (blue and cyan) represents the CRFs for cells contributing to recurrent inhibition. Higher contrast in the color indicates a stronger excitatory or inhibitory effect on the target cell. The black bar represents the target cell CRF. Note that as the bar length increases, the suppressive effect is mostly due to recurrent inhibition from cells that are a better description of the new stimulus (and therefore would be a more efficient stimulus description according to the sparse coding model).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3757072&req=5

pcbi-1003191-g009: Decomposition of the recurrent inputs contributing to the end-stopping effect.(A) Overall decomposition of the response into recurrent excitatory, inhibitory, and feedforward components; (B) Locations and orientations of the CRFs of cells contributing to the recurrent excitatory and inhibitory signals at different bar lengths. Only CRFs with significant influences are displayed (i.e., at steady state). The warmer color (yellow) represents the location and orientation of the CRFs for cells contributing to recurrent excitation, the cooler color (blue and cyan) represents the CRFs for cells contributing to recurrent inhibition. Higher contrast in the color indicates a stronger excitatory or inhibitory effect on the target cell. The black bar represents the target cell CRF. Note that as the bar length increases, the suppressive effect is mostly due to recurrent inhibition from cells that are a better description of the new stimulus (and therefore would be a more efficient stimulus description according to the sparse coding model).
Mentions: The recurrent interactions between cells in the sparse coding model implement a rich nonlinear response where cells compete to represent stimulus features. While it has been noted that stimuli in the CRF surround can produce sparse responses [5], [6], [8], the surprising finding of this work is that the particular form of inhibition and excitation necessary to implement a sparse coding model is sufficient to explain so many individual and population nCRF properties. At a high level, these effects likely arise from the present model because the observed responses produce a more efficient representation of the stimulus than alternative population responses. While a detailed investigation of how the network interactions give rise to the response properties is an interesting open question for future investigation, in general this is difficult to determine due to the interactions between the network dynamics and the stimulus dynamics (i.e., the response properties arise from the average response over a drifting grating, in addition to being influenced by network dynamics). In the case of end-stopping, the stimuli is not drifting and we can see more explicitly how this effect arises from the principles of sparse coding. In response to a given fixed stimuli, the steady-state network response is composed of a combination of feedforward excitation, recurrent excitation and recurrent inhibition. When plotting these three components of the steady-state response as a function of the bar length (Fig. 9A), it is evident that the overall response is mostly driven by the feedforward component and the recurrent inhibition. The feedforward excitation saturates as a result of the stimulus growing out of the CRF, but the recurrent inhibition keeps growing with increased bar length. To see the spatial extent of the recurrent influence, Fig. 9B shows the CRF locations and orientations of the cells influencing the target cell. As expected, inhibition mostly comes from cells with overlapping and co-linear CRFs that represent a more efficient description of the stimulus as the bar length increases.

Bottom Line: Models seeking to explain these non-classical receptive field (nCRF) effects in terms of circuit mechanisms, input-output descriptions, or individual visual tasks provide limited insight into the functional significance of these response properties, because they do not connect the full range of nCRF effects to optimal sensory coding strategies.Furthermore, when the population diversity of an nCRF effect has also been reported in the literature, we show that this model produces many of the same population characteristics.These results show that the sparse coding hypothesis, when coupled with a biophysically plausible implementation, can provide a unified high-level functional interpretation to many response properties that have generally been viewed through distinct mechanistic or phenomenological models.

View Article: PubMed Central - PubMed

Affiliation: Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States of America.

ABSTRACT
Extensive electrophysiology studies have shown that many V1 simple cells have nonlinear response properties to stimuli within their classical receptive field (CRF) and receive contextual influence from stimuli outside the CRF modulating the cell's response. Models seeking to explain these non-classical receptive field (nCRF) effects in terms of circuit mechanisms, input-output descriptions, or individual visual tasks provide limited insight into the functional significance of these response properties, because they do not connect the full range of nCRF effects to optimal sensory coding strategies. The (population) sparse coding hypothesis conjectures an optimal sensory coding approach where a neural population uses as few active units as possible to represent a stimulus. We demonstrate that a wide variety of nCRF effects are emergent properties of a single sparse coding model implemented in a neurally plausible network structure (requiring no parameter tuning to produce different effects). Specifically, we replicate a wide variety of nCRF electrophysiology experiments (e.g., end-stopping, surround suppression, contrast invariance of orientation tuning, cross-orientation suppression, etc.) on a dynamical system implementing sparse coding, showing that this model produces individual units that reproduce the canonical nCRF effects. Furthermore, when the population diversity of an nCRF effect has also been reported in the literature, we show that this model produces many of the same population characteristics. These results show that the sparse coding hypothesis, when coupled with a biophysically plausible implementation, can provide a unified high-level functional interpretation to many response properties that have generally been viewed through distinct mechanistic or phenomenological models.

Show MeSH
Related in: MedlinePlus