Hierarchical Predictive Coding Optimization via Precision-Weighted Bayesian Inference: A Computational Framework for Decomposing Top-Down Attentional Modulation in Working Memory Consolidation

Abstract

Contemporary models of working memory consolidation remain insufficiently equipped to account for the dynamic interplay between top-down attentional modulation and precision-weighted prediction error signaling. This study introduces a hierarchical Bayesian computational framework that systematically decomposes the contribution of cortical precision weighting to mnemonic encoding efficiency across four attentional load conditions (N = 218). Employing high-density electroencephalography (hdEEG) alongside drift-diffusion modeling and variational free-energy minimization protocols, we demonstrate that attentional gain modulates prediction error amplitude at distinct hierarchical levels of the prefrontal-parietal network. Results indicate that precision-weighted inference accounts for 61.4% of variance in consolidation fidelity, significantly outperforming classical resource-depletion models. These findings establish a neurobiologically grounded optimization schema for understanding adaptive memory encoding under cognitive load.