A deep learning approach to self-prioritization processing.

Abstract

Self-prioritisation is a ubiquitous psychological phenomenon which occurs when saliency-related processing is linked to self-related information compared to other-related information. Recent research has identified certain brain regions activated during self-saliency processing. However, the dynamic characteristics of the region connections causing such prioritization effects remain unknown. To address this, here we investigated neural couplings at the whole brain level during the self-saliency processing when participants carried out a standardised shape-label matching task widely used to test selfprioritization effects, while electroencephalogram (EEG) data were recorded. Behavioural results indicated the presence of the self-prioritization effect, evident by faster and more accurate responses to shape-label pairs associated with oneself compared to those related to a friend or a stranger. We then applied deep learning models to test and validate the robustness of the two phases of neural couplings. Specifically, in the early top-down regulation phase, the analysis based on SqueezeNet showed higher accuracy performance for the self-related stimuli compared to stimuli associated with others, but this is not the case for the later feedforward. Moreover, the deep learning models performed better in classifying the self-related stimuli in the early top-down phase than in the later feedforward phase, which was not detected in classifying stranger-related stimuli. The data indicated that deep learning networks might offer insights into psychological functioning that would otherwise be unresolved using traditional neuroscience methods; as a result, the latter might aid the development of brain-inspired algorithms for deep learning.