BabyBayes · Bayesian Learning in the Infant Brain

Bayesian Learning in the Infant Brain

The unique cognitive abilities observed in humans do not suddenly emerge in adulthood, but are rather deeply rooted in early infancy. Over the past decades, research in developmental psychology has repeatedly evidenced infants’ amazing learning abilities, dispelling the original vision of a neurologically insufficient and cognitively confused infant. However, developmental psychology still lacks an algorithmic understanding of early learning mechanisms, integrating cognitive computations with brain processes. Crucially, this knowledge gap has been holding the field back from having the resources to design theory-driven screening procedures with young infants in order to identify early markers of atypical development, before these deficits lead to profound delays. The development of non-invasive brain imaging techniques has recently opened the black-box of the infant brain, and, we are now at a turning point where we can move from the description of what infants can learn to the exploration of how the typically developing brain implements its learning strategies. The present project aims at bringing together developmental psychology, neuroscience and computational modelling to unravel the brain mechanisms underlying typically developing infants’ learning skills. The specific objectives were (1) to learn fNIRS, a neuroimaging technique compatible with infant research, and (2) to explore predictive coding in the infant brain, a neural coding strategy. During the implementation of this project, I learnt fNIRS during the outgoing phase, under the supervision of Richard Aslin. Due to important delays in the recruitment of infant participants, I primarily implemented this technique with adult participants, exploring the neural signature of speech and face processing. I was able to evidence clear neural responses to both stimuli from adult participants. The manuscript is under preparation. Due to the COVID crisis, I was unable to use this technique with infant participants, and I was forced to put the project on hold during that time. Instead, I wrote a review paper on steady-state evoked potentials, an EEG analysis technique particularly relevant with infant populations. This paper was published in 2022. During the return phase of the project, I implemented the EEG experimental part of the project, exploring predictive processes in the infant brain. I tested a total of 58 4-month-old infants, and I was able to include 36 of these infants in the study. Infants were presented with audio-visual events, with some auditory cues predicting the appearance of an image 1 sec after on the screen. The analysis of the recoded EEG activity indicate that infants quickly learn these audio-visual associations. Besides, the analyses revealed enhanced visual activity in response to unexpected vs expected visual omissions. Overall, these results indicate that predictive coding is a neural coding strategy available to the infant brain very early on.

Data: CORDIS, © European Union