Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster C in Poster Session C: Wednesday, August 5, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Changes in visual objectives explain differences in cue weighting across development: Simulations with neural network models

Olivia Maltz¹, Vladislav Ayzenberg¹; ¹Temple University

Presenter: Olivia Maltz

From a continuous visual landscape humans can extract individual, identifiable objects. Prior work has shown that when looking at a visual scene infants prioritize the spa- tial properties of objects, such as their locations, while preschoolers prioritize the featural properties of objects, such as their color or shape. We hypothesized that this developmental shift in cue weighting is driven by a transition from individuating objects, for which spatial cues are more informative, to identifying objects, for which featural cues are more informative. To test this hypothesis, we trained convolutional neural network (CNNs) models to either individuate or identify objects, and then tested models on the same cue weighting task as children. Like infants, the individuation model prioritized spatial cues, whereas, like preschoolers, the identification model prioritized featural cues. These results suggest that different visual objectives across development may lead to differences in how visual information is extracted and weighted.

Topic Area: Development, Individual Differences & Clinical Populations