Date created: | Last Updated:

Category: Project

Description: When we are exposed to a large number of scenes, we are remarkably good at memorizing them. The precise nature of large memory capacity is still an unresolved issue. Because visual scenes involve complex stimuli, it was difficult, until recently, to evaluate the similarity of two scenes. Researchers required a crucial manipulation check to decode this remarkable memory performance. One possible tool for helping us express how two scenes are perceptually similar are deep neural networks. In this study, we explored the extent to which we can create false visual memories using tasks inspired by the Deese-Roediger-McDermott paradigm. The similarity of scenes was evaluated by creating similarity space defined by a pre-trained deep network. The distances defined by the deep network were evaluated in a simple validation experiment. Results showed increased false alarm rates for scenes close to the averaged representation of the presented stimuli. A similar approach can be used for further studies regarding visual memory for complex scenes.