Interaction of contexts in context-dependent orientation estimation

Abstract

The processing of a visual stimulus is known to be influenced by the statistics in recent visual history and by the stimulus' visual surround. Such contextual influences lead to perceptually salient phenomena, such as the tilt aftereffect and the tilt illusion. Despite much research on the influence of an isolated context, it is not clear how multiple, possibly competing sources of contextual influence interact. Here, 10 using psychophysical methods, we compared the combined influence of multiple contexts to the sum of the isolated context influences. The results showed large deviations from linear additivity for adjacent or overlapping contexts, and remarkably, clear additivity when the contexts were sufficiently segregated.Specifically, for adjacent or overlapping contexts, the combined effect was often lower than the sum of the isolated component effects (sub-additivity), or was more influenced by one component than another 15 (selection). For contexts that were segregated in time (600 ms), the combined effect measured the exact sum of the isolated component effects (in degrees of bias). These results imply that the "encoding" of the context is a compressive non-linear transformation, followed by a non-linear selection, manifested by strong interactions between the encoded components. For selected encoded representations, the contextual influence is a linear sum of the individual effects, as predicted from a theory of optimally 20 combining reference cues.

A decision-time account of individual variability in context-dependent orientation estimation

“…We noted that behavioral biases such as TAE and TI may reflect a sum over different underlying mechanisms (Bao & Engel, 2012;Dekel & Sagi, 2019a). Possibly not all bias mechanisms are RT dependent.…”

“…Generally, contextdependent changes in visual processing are thought to be functionally useful, despite some debate regarding details (Clifford, 2014;Kohn, 2007;Snow, Coen-Cagli, & Schwartz, 2017;Solomon & Kohn, 2014;Webster, 2011). Possible benefits include (a) self-calibration, constancy, or correction of a reference "norm" (Andrews, 1964;Day, 1972;Dekel & Sagi, 2019a;Gibson & Radner, 1937;Webster, 2011), (b) optimization of the neural code, such as improved gain of computational units, improved coding sensitivity to likely events, or decorrelation to remove coding redundancies (Benucci, Saleem, & Carandini, 2013;Coen-Cagli, Kohn, & Schwartz, 2015;Pinchuk-Yacobi & Sagi, 2019;Snow et al, 2017;Wei & Stocker, 2017), and (c) enhanced attentional selection of novel or surprising events (such events are presumably more likely to be important and hence deserve more attention). However, these and other alternatives are not necessarily mutually exclusive (e.g., orientation biases may reflect both self-calibration and decorrelation, Clifford, Wenderoth, & Spehar, 2000), and are not necessarily dependent on the neural implementation (e.g., divisive normalization may underlie both code optimization and attentional selection, Carandini & Heeger, 2012).…”

A decision-time account of individual variability in context-dependent orientation estimation

“…We noted that behavioral biases such as TAE and TI may reflect a sum over different underlying mechanisms (Bao & Engel, 2012;Dekel & Sagi, 2019a). Possibly not all bias mechanisms are RT 535…”

“…Generally, contextdependent changes in visual processing are thought to be functionally useful, despite some debate 40 regarding details (Clifford, 2014;Kohn, 2007;Snow, Coen-Cagli, & Schwartz, 2017;Solomon & Kohn, 2014;Webster, 2011). Possible benefits include (a) self-calibration, constancy, or correction of a reference "norm" (Andrews, 1964;Day, 1972;Dekel & Sagi, 2019a;Gibson & Radner, 1937;Webster, 2011), (b) optimization of the neural code, such as improved gain of computational units, improved coding sensitivity to likely events, or decorrelation to remove coding redundancies (Benucci, Saleem, & & Webster, 2017). In the TI, as well as in other spatial-context-dependent biases, individuals measure large differences (by an order of magnitude), with strong test-retest reliability (Grzeczkowski et al, 2017;Song, Schwarzkopf, & Rees, 2013).…”