Triple Configuration of Brain Networks Based on Recurrent Neural Networks: The Synergistic Effects of Exogenous Stimuli, Task Demands, and Spontaneous Activity

Binghao Yang, Guangzong Chen · 2026

Read Triple Configuration of Brain Networks Based on Recurrent Neural Networks: The Synergistic Effects of Exogenous Stimuli, Task Demands, and Spontaneous Activity by Binghao Yang, Guangzong Chen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-23525.txt.

Introduction

Yang and Chen model how brain nets switch setups under outside cues, task demands, and the brain at rest, using recurrent nets shaped by neural dynamics. Fitting resting-state EEG data from 114 people, they find the parietal net acts as the hub for these patterns. Front and back parietal zones play distinct roles across cue types, the model shows, showing the parietal role in nimble thinking.

Catalog record

CREATORBinghao Yang, Guangzong Chen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39643
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.23525

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Triple Configuration of Brain Networks Based on Recurrent Neural Networks: The Synergistic Effects of Exogenous Stimuli, Task Demands, and Spontaneous Activity Binghao Yang, Guangzong Chen Abstract The foundation of cognitive flexibility and higher-order intelligence lies in the functional structure and activity of brain networks, which can be dynamically configured by both external environments and internal states. However, decoding these dynamics from high-dimensional neural data remains a challenge. In this study, we propose a computational framework using Recurrent Neural Networks (RNNs) with neural dynamic constraints to model source-localized resting-state EEG data from 114 114 participants. We aim to clarify the “triple brain network configurations” driven by exogenous and endogenous factors, including external stimuli, information processing tasks, and spontaneous activities. Our model identifies the parietal network as a critical hub supporting these multiple configuration patterns. Furthermore, we reveal that the anterior and posterior parietal regions exhibit distinct functional specializations under different stimulus modalities. By formalizing a triple configuration framework, this work separates latent factors of brain dynamics and underscores the computational significance of parietal regions in orchestrating higher-order intelligence. 1 Introduction As a dynamic system, the brain continuously receives and processes external information while performing multiple cognitive functions [ 4 , 35 ] . It exhibits highly complex activity patterns for encoding and processing information [ 11 ] , and can flexibly transition between distinct cognitive tasks and contexts [ 31 , 30 , 22 ] . Such cognitive processing relies on the recruitment of distributed brain regions, effectively manifesting as the dynamic configuration of brain networks [ 38 ] . The ability to selectively recruit brain regions to meet varying cognitive demands serves as the foundation of the brain’s complexity and flexibility. Specifically, the configuration of brain networks is crucial for the processing of complex information and higher-order cognitive functions across multiple domains, including sensory processing, attention, learning, and memory. For instance, interactions between the insula and temporal networks enable the brain to decompose information within complex auditory environments [ 3 , 8 ] . Similarly, attention modulates visual network configuration to amplify the encoding of specific features in visual input [ 16 ] , while learning processes in uncertain environments involve the reconfiguration of the frontoparietal network [ 24 , 23 ] . Furthermore, the configuration within the limbic network—specifically the interaction between the hippocampus and the amygdala—supports memory processing distinctively during the encoding and retrieval phases [ 6 , 20 ] . However, several key questions remain elusive. Given the complexity of cognition, network configuration is influenced by multifarious factors. For example, the concurrent presence of primary sensory responses and higher-order cognitive processing following task-related stimuli makes it difficult to disentangle their respective contributions to network configuration. Isolating these factors is essential for clarifying the specific roles of brain networks in stimulus response versus cognitive processing. Additionally, in natural environments, brain networks must accommodate multiple inputs and tasks, a process further modulated by the brain’s intrinsic state [ 38 ] . The network mechanisms enabling adaptation to such diverse demands, as well as their relationship with spontaneous brain activity, remain unclear. In this study, by employing recurrent neural networks (RNNs) to model neural dynamics from resting-state EEG and subjecting these biologically constrained models to various external inputs and cognitive tasks, we propose a “triple configuration framework” governing brain networks via exogenous and endogenous effects. For a detailed description of the study pipeline and the triple configuration framework, refer to Appendix A (Figure 6 ). 2 Materials and Methods 2.1 Participants, EEG Recording, Preprocessing, and Source Localization We recruited 114 participants for this study. EEG data were recorded using a 64 -channel system and preprocessed using standard pipelines, including filtering and Independent Component Analysis (ICA) for artifact removal. Detailed information regarding participants, data acquisition, preprocessing, and source localization is provided in Appendix A. 2.2 Recurrent Neural Network Model Training Figure 1: Stimulus-driven network configuration under auditory cortex input. (A) Functional connectivity patterns under white noise input with bilateral correlation 1. Upper-Left: pairwise whole brain connectivity; Upper-right: sub-network connectivity; Lower-left: averaged seed-point connectivity of sub-networks. Lower-right: seed-point connectivity mapping on the whole brain. (B) Same as (A) under white noise input with bilateral correlation 0. (C) Same as (A) under 40 Hz input. (D) Network configuration by the change of correlated characteristic (Noise correlation 1 vs Noise correlation 0). Left: seed-point connectivity configuration of sub-networks; Middle: seed-point connectivity configuration mapping on the whole brain; Right: pairwise connectivity configuration of the whole brain. (E) Same as (D) by the change of periodic characteristic (Noise correlation 1 vs 40 Hz). (F) Left: Stimulus-driven common network by combining (D) and (E); Upper-right: Network only configured by the change of correlated characteristic; Lower-right: Network only configured by the change of periodic characteristic. (G) Same as (F) shown at the pairwise connectivity level. * p<0.05 after Bonferroni correction. Figure 2: Task-driven network configuration under auditory cortex input. (A) Network configuration performing integration task under the input of white noise with correlation 0. Left: seed-point connectivity configuration of sub-networks; Right: seed-point connectivity configuration mapping on the whole brain. (B) Same as (A) under the input of 40 Hz with phase lag 90^{\circ} . (C) Network configuration performing separation task under the input of white noise with correlation 0. Left: seed-point connectivity configuration of sub-networks; Right: seed-point connectivity configuration mapping on the whole brain. (D) Same as (C) under the input of 40 Hz with phase lag 90^{\circ} . (E) Upper: Task-driven common network by combining (A) - (D); Lower-left: Network only configured by the integration task; Lower-right: Network only configured by the separation task. * p<0.05 after Bonferroni correction. Figure 3: Spontaneous-activity network configuration under auditory cortex input. (A) Schematic of similarity analysis, i.e. calculating the correlation coefficient between stimulus-driven functional connectivity and resting-state functional connectivity. (B) Similarity analysis across five frequency bands: delta ( 1-4 Hz), theta ( 4-8 Hz), alpha ( 8-12 Hz), beta ( 12-30 Hz), and gamma ( 30-45 Hz). Upper: Similarity with functional connectivity under white noise input. Lower: Similarity with functional connectivity under 40 Hz input. (C) Stimulus-task common network combining Figure 1 (F) and Figure 2 (E). (D) Correlation of common network configuration patterns. Upper: different types of stimulus input; Lower: different types of tasks. (E) Comparison between the degree of brain regions in the common network and the whole-brain averaged degree. Upper: alpha band; Lower: beta band. * p<0.05 after Bonferroni correction; *** p<0.001 . Figure 4: Elimination experiment. (A) Change of the model performance (measured by the correlation coefficient between model output signal and target signal) under control condition (without elimination), random-disturbed condition (eliminating read-out weights randomly) and common network disturbed condition (eliminating read-out weights corresponding with the common network) in the integration task. (B) Same as (A) in the separation task. *** p<0.001 after Bonferroni correction; Figure 5: Triple network configuration under visual cortex input. (A)-(D) Stimulus-driven configuration. (A) Network configuration by the change of correlated characteristic (Noise correlation 1 vs Noise correlation 0). Left: seed-point connectivity configuration of sub-networks; Middle: seed-point connectivity configuration mapping on the whole brain; Right: pairwise connectivity configuration of the whole brain. (B) Same as (A) by the change of periodic characteristic (Noise correlation 1 vs 40 Hz). (C) Left: Stimulus-driven common

Read the complete text