Process reveals structure: How a network is traversed mediates expectations about its architecture

Abstract

Network science has emerged as a powerful tool through which we can study the higher-order architectural properties of the world around us. How human learners exploit this information remains an essential question. Here, we focus on the temporal constraints that govern such a process. Participants viewed a continuous sequence of images generated by three distinct walks on a modular network. Walks varied along two critical dimensions: their predictability and the density with which they sampled from communities of images. Learners exposed to walks that richly sampled from each community exhibited a sharp increase in processing time upon entry into a new community. This effect was eliminated in a highly regular walk that sampled exhaustively from images in short, successive cycles (i.e., that increasingly minimized uncertainty about the nature of upcoming stimuli). These results demonstrate that temporal organization plays an essential role in learners' sensitivity to the network architecture underlying sensory input.

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Figure 1

Representation of the graph and walk structure underlying visual sequences. The graph consisted of three distinct communities of interconnected nodes (shown in yellow, teal, and purple). Each node in the graph corresponded to a unique fractal image, and edges between nodes corresponded to their possible co-occurrence in a sequence. Sequences were generated by “walking” along the edges of the graph randomly, or according to successive Eulerian and Hamiltonian paths. In the color-coded walk samples shown above, we illustrate that sequences generated by Random walks and Eulerian paths tended to stay within a given community (relative to Hamiltonian paths, which only sparsely sampled from each community).

Figure 2

Boxplots of reaction time increases across experimental conditions (N = 59). Cross-community surprisal effects were calculated by subtracting, for each participant, mean RTs for pre-transition nodes from mean RTs for transition nodes. A value greater than 0 indicates an increase in RT upon entry into a new community during the exposure phase. Note that strong evidence for surprisal is observed only for walk types involving repeated exposure to common connections within the same community (Eulerian and Random). No surprisal effect was observed for participants in the Hamiltonian condition.