Continuous action with a neurobiologically inspired computational approach reveals the dynamics of selection history
Fig 3
An overview of SH-CoR architecture.
The model consists of two processes: Target Selection (left) and Movement Production (right). Each process comprises several modules (boxes with solid black lines). The Target Selection process consists of layers of model neurons (illustrated dotted lines, as circles C1, C2, C3, and C4 for four colors used in the experiment) in the modules, modeling the neuron-like responses to the four colors. The shading of the nodes illustrates the level of activation for the current trial, with white being the highest activation and black the lowest. Both processes and their modules operate in parallel (analogous to brain regions), allowing SH-CoR to produce the leakage effect. The first stage of Target Selection (Color Processing module) determines color Feature Maps and the saliency of the colors present in the Color Display using different color units (Color Saliency Layer). For instance, in the above display, C1 (white circle) and C2 (gray) represent the most and least salient colors in the display, while C3 and C4 (black circles) represent colors absent in the display. Based on the output of the Color Saliency Layer (see S3.A and Eq. S1 in S3 Text), the Odd Color Selection module selects the most salient color. It identifies the distractor and absent colors via color competition (Eq. S2 in S3 Text). These feature units are assigned labels, T (target color), D (distractor color), and N (absent color). The line graphs above the units remind us that color competition is a temporal process (not instantaneous) whose speed is proportional to the color saliency. These temporal activations are combined with the output of the Feature Maps in a multiplicative way (Eq. S3 in S3 Text). Initially, all possible item locations compete to become the target location, but the color competition ensures that the salient item dominates the location competition. This way, the feature map of the winning color (odd-color) eventually dominates the input to a competition of locations, generating the Target Location Representation (Eq. S4 in S3 Text). In addition, the Target Selection is also influenced by the Selection History module. This process stores the target and distractor features from the previous trial (see dotted lines; Eq. S6 in S3 Text) in separate layers. The layers in the Selection History module, in turn, influence the selection of the current target through facilitation and/or inhibition mechanism depending on which of the five models of inter-trial selection history are implemented in a particular instantiation of SH-CoR (see the gray dialog box).