文献类型：期刊文献

英文题名：Implementing general working memory via Hebbian plasticity in a theta-gamma coupled network

作者：Liu, Dongyan[1];Xu, Xuying[1,2];Wang, Yihong[1,2];Pan, Xiaochuan[1,2];Du, Ying[1,2];Wang, Rubin[1,3]

机构：[1]East China Univ Sci & Technol, Inst Cognit Neurodynam, Sch Math, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Sch Math, Ctr Intelligent Comp, Shanghai 200237, Peoples R China;[3]Hangzhou Dianzi Univ, Sch Comp Sci, Hangzhou 200237, Peoples R China

年份：2026

卷号：680

外文期刊名：NEUROCOMPUTING

收录：;EI(收录号：20261220293292);WOS:【SCI-EXPANDED(收录号:WOS:001730403100001)】；

基金：This work was supported by the National Natural Science Foundation of China (Nos. 12272136, 12172132 and 12472054) and Science and Technology Commission of Shanghai Municipality (No. 24JS2810400).

语种：英文

外文关键词：Working memory; Hebbian plasticity; Theta-gamma coupling; Neural network model

摘要：Working memory, a core cognitive system for information processing and temporary storage, is widely believed to rely on coordinated neural oscillations. Theta-gamma coupling has been identified as a key mechanism for the temporal encoding of multiple memory items. However, existing computational models are predominantly specialized for single tasks, lacking a unified framework to explain the general principles underlying diverse working memory functions. To address this, we developed a biophysically grounded, multi-layer neural network model that incorporates theta-gamma coupling and is trained via multi-to-multi Hebbian plasticity rules. This model successfully performs three distinct types of cognitive tasks: associative memory, by recalling complete objects from partial cues and enabling cross-object associations through feature overlap; sequential memory, by identifying the most recently learned item among distractors and continuing to recall subsequent sequences, while also flexibly switching task rules in multi-cue paradigms; and spatial navigation memory, by executing both forward and reverse replay of learned paths. Crucially, we demonstrate that lesioning the entorhinal-hippocampal pathway disrupts theta-gamma phase-amplitude coupling and leads to a complete failure in sequential recall. Furthermore, perturbing the intrinsic excitation-inhibition balance within cortical columns degrades theta-gamma oscillations and impairs path replay. Our work provides computational evidence that a unified theta-gamma coding mechanism, shaped by Hebbian plasticity, can support a spectrum of working memory tasks across cognitive domains, offering a unified framework for understanding their core computational and neural principles.