Jean-Rémy HOCHMANN^1,2 (), Céline Spriet^1,2, Liuba PAPEO^1,2; ¹CNRS UMR5229 – Institut des Sciences Cognitives Marc Jeannerod, 67 Bd Pinel 69675 Bron, France, ²Université Claude Bernard Lyon 1, France

The hierarchical neuronal and cognitive underpinning of visual perception are largely in place early in life, possibly at birth, but immature brain and fledgling myelination might slow down the transmission and integration of information into complex representations. In 5 experiments involving 160 infants, we used frequency-tagging electroencephalography combined with rapid serial visual presentation to show that the capacity to categorize objects at 4 and 9 months is respectively seven times and three times slower than in adults. Next, we tested adults with accelerated visual streams to simulate object perception in young infants. We used representational similarity analysis (RSA) to compare representational models extracted respectively from a multivariate pattern analysis of the adult EEG signal and from the eight layers of an artificial deep convolutional neural network (CNN) trained on object categorization (Alexnet). RSA revealed significant correlations between EEG-based representations and each layer of the CNN, for all speed conditions (6, 12 or 30 Hz). Cluster-mass permutation tests however revealed that increased speed of presentation from 12 and 30 Hz reduced correlations of neural representations with layers conv3, conv4 and conv5 around 205-314 ms. This suggests that, when visual streams accelerated, low complexity features, accessed early and coded in early layers modelling EVC, were maintained, whereas higher-complexity visual features akin to those computed in more anterior visual areas were partly lost. In other words, when streams accelerated, lower-level sensory information was mostly preserved, but the computations transforming that information into meaningful categories were impacted. Taking adults’ visual perception at faster stimulus presentation as a simulation of young infants’ perception, these results suggest that infants may have access to more limited feature maps, yielding representation of less detailed, broader categories. Slow vision may constitute a design feature of early development, highlighting the most stable and general features of the world.

Acknowledgements: L. P. was supported by a European Research Council Starting Grant (Project: THEMPO, Grant Agreement 758473). J.-R. H. was supported by the Fondation de France (Grant number: 00134704 / WB-2022-46104).