Jin Fan

This study evaluates the limitations of transformer attention mechanisms in large language models by testing their performance on a Stroop task, revealing degraded conflict resolution with longer sequences.

This study evaluates the ability of three large language models to identify and describe emotions, finding that GPT 4 performs close to human levels while Bard and GPT 3.5 show lower performance.

The paper uses machine learning to identify morphometrical brain markers of sex differences in adolescents, finding specific brain regions that distinguish males and females and are linked to psychopathology.

The paper investigates supramodal executive control of attention by examining conflict processing in unimodal and crossmodal dual-conflict tasks, finding additive interference effects under time constraints.

This study demonstrates that the anterior insular cortex is critical for interoceptive attention by showing its activation and connectivity patterns in healthy individuals and impaired function in patients with AIC lesions.

This study shows right hemisphere superiority in executive control of attention, dependent on alerting and orienting functions, using a modified Attentional Network Test.