Search Papers | Poster Sessions | All Posters
Poster C89 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Efficient Neural Compression of Sensory Information during Human Category-Based Decisions
Julie Drevet1 (), Jan Drugowitsch2, Valentin Wyart1; 1Ecole normale supérieure Paris - INSERM, 2Harvard Medical School, Department of Neurobiology
Human visual categorization relies on an inference process that extracts the statistics of ambiguous sensory observations through imprecise computations. But theories diverge regarding whether this imprecise inference process integrates sensory information in its native stimulus space or in a compressed space centered on decision-relevant categories. Here we designed a visual categorization task in which we manipulated the space in which human observers can perform inferences. We found that humans perform more accurate inferences when integrating sensory information in category space. Concurrent magnetoencephalographic recordings showed accuracy-predictive signatures of compressed neural representations of sensory information in conditions where humans can perform inferences in category space. Together, these findings indicate that humans mitigate the costs of imprecise inferences by focusing limited computational resources on decision-relevant information.
Keywords: human inference magnetoencephalography visual categorization computational model