Investigating attention manipulation in behavioural data using Hierarchical Gaussian Filter and neural network models
Abdullah Al Saqib Majumder · 2024
Attention is a critical cognitive process that has been hypothesized to enable the brain to selectivelyfocus on relevant stimuli while filtering out distractions, thereby optimizing the allocation of limitedcomputational resources. Within the predictive coding framework, attention is conceptualized as amechanism that enhances the precision of sensory predictions, thereby reducing prediction errorsand improving cognitive efficiency. This thesis investigates attention manipulation in behavioral datathrough the application of the Hierarchical Gaussian Filter (HGF) and neural network models, with afocus on how diverted attention affects precision, and therefore surprise optimization, and difficultyin perceptual tasks. Using the HGF framework, we modeled hierarchical inference processes, exam-ining how tonic volatilities of the higher levels as well as expected precisions and expected meansof the models capture attentional differences under full and diverted attention conditions. The studyalso introduced contrast as a proxy for task difficulty, providing a quantifiable measure of the cognitive load associated with diverted attention. The neural network models were employed to explore weight distributions and saliency maps, offering a complementary perspective on how attention influences neural representations. While the neural network was able to identify substantial differences in the two attention conditions, albeit at the group level, the results from HGF models of indivindual participants revealed only a weak correlation between model parameters and actual attentional differences, suggesting that the models struggled to fully capture the complexity of attentional shifts between the two conditions. The findings underscore the challenges in using computational models to infer cognitive processes like attention, particularly when dealing with subtle variations in task difficulty and attention diversion.