Limits...
Nonlinear circuits for naturalistic visual motion estimation.

Fitzgerald JE, Clark DA - Elife (2015)

Bottom Line: Furthermore, a diversity of inputs to motion detecting neurons can provide access to more complex higher-order correlations.Collectively, these results illustrate how non-canonical computations improve motion estimation with naturalistic inputs.This argues that the complexity of the fly's motion computations, implemented in its elaborate circuits, represents a valuable feature of its visual motion estimator.

View Article: PubMed Central - PubMed

Affiliation: Center for Brain Science, Harvard University, Cambridge, United States.

ABSTRACT
Many animals use visual signals to estimate motion. Canonical models suppose that animals estimate motion by cross-correlating pairs of spatiotemporally separated visual signals, but recent experiments indicate that humans and flies perceive motion from higher-order correlations that signify motion in natural environments. Here we show how biologically plausible processing motifs in neural circuits could be tuned to extract this information. We emphasize how known aspects of Drosophila's visual circuitry could embody this tuning and predict fly behavior. We find that segregating motion signals into ON/OFF channels can enhance estimation accuracy by accounting for natural light/dark asymmetries. Furthermore, a diversity of inputs to motion detecting neurons can provide access to more complex higher-order correlations. Collectively, these results illustrate how non-canonical computations improve motion estimation with naturalistic inputs. This argues that the complexity of the fly's motion computations, implemented in its elaborate circuits, represents a valuable feature of its visual motion estimator.

Show MeSH
Separate ON and OFF processing improved motion estimation by supplementing the HRC with odd-ordered correlations.(A) By summing and subtracting the four quadrants (top labels, e.g., ‘++’) in four different patterns, we isolated the contributions of various correlation types (side labels, e.g., ‘odd’) to the weighted 4-quadrant model (Appendix 7). For example, the uniform sum of the four quadrants is the HRC, and we denote this quadrant combination as ‘even = 2’ (top row of matrix). The other three rows of the matrix define quadrant combinations that are sensitive to two different classes of third and higher odd-ordered correlations (‘odd’ and ‘odd*’ rows) and to fourth and higher even-ordered correlations (‘even >2’ row). The factor of 1/4 merely sets the magnitude of the quadrant contributions to match the formulas in Appendix 7 and is without conceptual importance. (B) The ‘even = 2’ correlation class worked best in isolation. Nevertheless, the ‘even = 2’ and ‘odd’ classes were highly synergistic (their weighted sum is notated ‘best 2’), and these classes together made the ‘odd*’ and the ‘even >2’ classes irrelevant.DOI:http://dx.doi.org/10.7554/eLife.09123.008
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4663970&req=5

fig3s1: Separate ON and OFF processing improved motion estimation by supplementing the HRC with odd-ordered correlations.(A) By summing and subtracting the four quadrants (top labels, e.g., ‘++’) in four different patterns, we isolated the contributions of various correlation types (side labels, e.g., ‘odd’) to the weighted 4-quadrant model (Appendix 7). For example, the uniform sum of the four quadrants is the HRC, and we denote this quadrant combination as ‘even = 2’ (top row of matrix). The other three rows of the matrix define quadrant combinations that are sensitive to two different classes of third and higher odd-ordered correlations (‘odd’ and ‘odd*’ rows) and to fourth and higher even-ordered correlations (‘even >2’ row). The factor of 1/4 merely sets the magnitude of the quadrant contributions to match the formulas in Appendix 7 and is without conceptual importance. (B) The ‘even = 2’ correlation class worked best in isolation. Nevertheless, the ‘even = 2’ and ‘odd’ classes were highly synergistic (their weighted sum is notated ‘best 2’), and these classes together made the ‘odd*’ and the ‘even >2’ classes irrelevant.DOI:http://dx.doi.org/10.7554/eLife.09123.008

Mentions: We next considered all subsets of two, three, or four quadrants. The best subsets for each number of predictors were nested, and the quadrants were incorporated in the order (i) (− −); (ii) (− +); (iii) (+ +); (iv) (+ −). Although all four quadrants enhanced the accuracy of the weighted 4-quadrant model, the benefit of each added quadrant decreased with the number of quadrants (Figure 3C). It is possible to reparameterize the weighted 4-quadrant model in a form that isolates the contributions of various higher-order correlations to the model's accuracy (Appendix 7). Interestingly, this parameterization showed that nearly all the accuracy of the weighted 4-quadrant model can be obtained by supplementing the HRC with a set of odd-ordered correlations that account for the asymmetry between positive and negative low-pass filtered signals (Figure 3—figure supplement 1, Appendix 8). Principal component analysis (PCA) did not reveal this simple interpretation of the model's computation (Appendix 9).


Nonlinear circuits for naturalistic visual motion estimation.

Fitzgerald JE, Clark DA - Elife (2015)

Separate ON and OFF processing improved motion estimation by supplementing the HRC with odd-ordered correlations.(A) By summing and subtracting the four quadrants (top labels, e.g., ‘++’) in four different patterns, we isolated the contributions of various correlation types (side labels, e.g., ‘odd’) to the weighted 4-quadrant model (Appendix 7). For example, the uniform sum of the four quadrants is the HRC, and we denote this quadrant combination as ‘even = 2’ (top row of matrix). The other three rows of the matrix define quadrant combinations that are sensitive to two different classes of third and higher odd-ordered correlations (‘odd’ and ‘odd*’ rows) and to fourth and higher even-ordered correlations (‘even >2’ row). The factor of 1/4 merely sets the magnitude of the quadrant contributions to match the formulas in Appendix 7 and is without conceptual importance. (B) The ‘even = 2’ correlation class worked best in isolation. Nevertheless, the ‘even = 2’ and ‘odd’ classes were highly synergistic (their weighted sum is notated ‘best 2’), and these classes together made the ‘odd*’ and the ‘even >2’ classes irrelevant.DOI:http://dx.doi.org/10.7554/eLife.09123.008
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4663970&req=5

fig3s1: Separate ON and OFF processing improved motion estimation by supplementing the HRC with odd-ordered correlations.(A) By summing and subtracting the four quadrants (top labels, e.g., ‘++’) in four different patterns, we isolated the contributions of various correlation types (side labels, e.g., ‘odd’) to the weighted 4-quadrant model (Appendix 7). For example, the uniform sum of the four quadrants is the HRC, and we denote this quadrant combination as ‘even = 2’ (top row of matrix). The other three rows of the matrix define quadrant combinations that are sensitive to two different classes of third and higher odd-ordered correlations (‘odd’ and ‘odd*’ rows) and to fourth and higher even-ordered correlations (‘even >2’ row). The factor of 1/4 merely sets the magnitude of the quadrant contributions to match the formulas in Appendix 7 and is without conceptual importance. (B) The ‘even = 2’ correlation class worked best in isolation. Nevertheless, the ‘even = 2’ and ‘odd’ classes were highly synergistic (their weighted sum is notated ‘best 2’), and these classes together made the ‘odd*’ and the ‘even >2’ classes irrelevant.DOI:http://dx.doi.org/10.7554/eLife.09123.008
Mentions: We next considered all subsets of two, three, or four quadrants. The best subsets for each number of predictors were nested, and the quadrants were incorporated in the order (i) (− −); (ii) (− +); (iii) (+ +); (iv) (+ −). Although all four quadrants enhanced the accuracy of the weighted 4-quadrant model, the benefit of each added quadrant decreased with the number of quadrants (Figure 3C). It is possible to reparameterize the weighted 4-quadrant model in a form that isolates the contributions of various higher-order correlations to the model's accuracy (Appendix 7). Interestingly, this parameterization showed that nearly all the accuracy of the weighted 4-quadrant model can be obtained by supplementing the HRC with a set of odd-ordered correlations that account for the asymmetry between positive and negative low-pass filtered signals (Figure 3—figure supplement 1, Appendix 8). Principal component analysis (PCA) did not reveal this simple interpretation of the model's computation (Appendix 9).

Bottom Line: Furthermore, a diversity of inputs to motion detecting neurons can provide access to more complex higher-order correlations.Collectively, these results illustrate how non-canonical computations improve motion estimation with naturalistic inputs.This argues that the complexity of the fly's motion computations, implemented in its elaborate circuits, represents a valuable feature of its visual motion estimator.

View Article: PubMed Central - PubMed

Affiliation: Center for Brain Science, Harvard University, Cambridge, United States.

ABSTRACT
Many animals use visual signals to estimate motion. Canonical models suppose that animals estimate motion by cross-correlating pairs of spatiotemporally separated visual signals, but recent experiments indicate that humans and flies perceive motion from higher-order correlations that signify motion in natural environments. Here we show how biologically plausible processing motifs in neural circuits could be tuned to extract this information. We emphasize how known aspects of Drosophila's visual circuitry could embody this tuning and predict fly behavior. We find that segregating motion signals into ON/OFF channels can enhance estimation accuracy by accounting for natural light/dark asymmetries. Furthermore, a diversity of inputs to motion detecting neurons can provide access to more complex higher-order correlations. Collectively, these results illustrate how non-canonical computations improve motion estimation with naturalistic inputs. This argues that the complexity of the fly's motion computations, implemented in its elaborate circuits, represents a valuable feature of its visual motion estimator.

Show MeSH