Behavioural and brain activation differences in Hierarchical Bayesian Inference across Age

1371 

Abstract Submission 

Vae Zhang¹, Wilson Lim¹, SH Annabel Chen²

¹School of Social Sciences, Psychology, Nanyang Technological University Singapore, Singapore, Singapore, ²School of Social Sciences, Psychology, Nanyang Technological University Singapore, Singapore, Not applicable

Vae Zhang, PhD Candidate  
School of Social Sciences, Psychology, Nanyang Technological University Singapore
Singapore, Singapore

Wilson Lim  
School of Social Sciences, Psychology, Nanyang Technological University Singapore
Singapore, Singapore

SH Annabel Chen, PhD  
School of Social Sciences, Psychology, Nanyang Technological University Singapore
Singapore, Not applicable

Hierarchical causal relations underlie state changes in dynamic environments[1]. These relations could be encoded in the brain hierarchically[2][3] and optimal inference occurs via Bayesian estimation of the context of incoming sensory evidence[4]. The neural correlates of Hierarchical Bayesian Inference (HBI) and their changes across age are not well elucidated. Thus, we examined the neural encoding of HBI across age groups in a probabilistic task during an EEG recording (fig.1).

30 older (age 55-72) and 30 younger adults (age21-34) were included in the analysis. Subject-wise HBI parameters were computed for each trial using the Hierarchical Gaussian Filter[4]. 64-channel EEG activity was preprocessed before computing each frequency band's relative power across epochs for all trials of each subject from different ROIs for the pre-and post-stimulus period (fig.1B). Correlations between HBI parameters and relative power were examined across trial types and age groups using linear models in R. All pairwise post-hoc comparisons were false discovery rate corrected.

Older adults had lower sensory surprise (p=0.002), higher variances in their estimation of contextual rules(p<0.001), and higher estimated variance in the mean rate of contextual changes (p=0.015) as compared to young adults. Before stimulus onset, a significant interaction between age group and trial type in the occipital region was found in the linear correlations between beta band power and sensory surprise. Opposite trends between younger and adults for these high-surprise trial correlations were significantly different (post-hoc p = 0.031). The linear correlation between beta power and estimated variance in sensory predictions was higher in high-surprise trials than in low-surprise trials in the parietal region (p=0.021). In the post-stimulus period, linear correlations between alpha band power and estimated variance in contextual rules were significantly higher in low surprise trials as compared to high surprise trials in the parietal region (p=0.007). There was a significant interaction effect between age and trial type for the linear correlation between alpha power and subjects' estimates of the mean rate of change of context in the prestimulus period. The linear correlation between the two variables was significantly higher in younger adults for the high surprise trials compared to old adults in the central regions of the brain (post-hoc p=0.0025).

Hierarchical Bayesian modelling revealed that older adults may experience lower than necessary levels of sensory prediction errors which resulted in comparatively higher variances in the estimation of contextual contingencies and thus higher variance in the estimation of context volatility, as compared to younger adults. Specific neural hierarchies that might be responsible for encoding HBI were found, and changes in these hierarchies may have resulted in different uncertainty estimates across ages. Posterior beta power in the prestimulus period increased with prediction error and estimated variance in predictions which is consistent with its proposed role in signalling predictive expectancies [5,6]. This relationship was inverted in older adults which may explain why older adults experience lower sensory surprise. Alpha power is proposed to correlate both negatively and positively with stimulus uncertainty[7,8]. These differences may be resolved by looking at uncertainty from a hierarchical perspective. We found that the correlations between poststimulus parietal alpha power correlated negatively with estimated precision and update of contextual rules in high surprise trials, while central prestimulus alpha power correlated positively with context volatility which is a higher level of uncertainty. This was inverted for older adults in high surprise trials which could suggest poorer inhibition control in the central region.

Executive Function, Cognitive Control and Decision Making

Reasoning and Problem Solving

Learning and Memory Other

Aging ²

Bayesian Modeling ¹

Aging

Cognition

Computational Neuroscience

Cortex

Electroencephaolography (EEG)

Learning

Modeling