Most existing open-loop Deep Brain Stimulation (DBS) systems for drug-resistant epilepsy do not take into account the temporal dynamics of seizures or the spread of seizure activity. In this paper, we propose a closed-loop control system that models seizure prevention as a constraint on the nonlinear dynamics of the brain’s structural network, and we evaluate it entirely in simulation. The structural network of the brain is modeled using the HCP-1065 (Human Connectome Project) connectome, which divides the brain into 360 regions. A densely connected sub-network of 60 brain regions was used to develop and test the proposed control system. Each region is modeled as a nonlinear supercritical Hopf oscillator, and the seizures modeled are focal-onset events that spread from a hippocampal onset zone through the connectome. A detector continuously monitors critical-slowing-down features of the dynamics, such as variance and lag-1 autocorrelation, to identify the approach to seizure activity. Two levels of control are applied: a linear-quadratic regulator (Level 1) that maintains the Hopf oscillators’ basal state, and a dormant network-wide optimal controller (Level 2) that employs an adjoint-based optimal-control method (Adjoint Optimal Control; AOC) to provide pre-emptive control. Within the model, the predictive detector identifies the approach to the seizure transition with an area under the receiver operating characteristic curve (AUC) of 0.889 (95
Understanding how collective neuronal rhythms emerge from the interplay between network structure and single-neuron dynamics remains a fundamental challenge in neurodynamics. In this work, we show that in small heterogeneous networks, the network topology and the position of a single tonic neuron govern synchronization dynamics. We study minimal four-neuron Huber–Braun motifs composed of one tonic neuron and three bursting neurons coupled via electrical synapses. By systematically varying the tonic neuron’s centrality—quantified through degree and a quasi-Laplacian measure—we identify two critical coupling thresholds: the critical coupling for global phase synchronization and the coupling at which the network returns to synchronized periodic bursting. Remarkably, centrality determines the nature of the synchronized attractor: motifs in which the tonic neuron has low centrality synchronize into tonic spiking, whereas highly central tonic neurons drive the network toward synchronized bursting. Moreover, the networks exhibit nontrivial collective behaviors, including large-ISI “stairway-to-heaven” structures, partial firing death characterized by the coexistence of spiking and subthreshold neurons, and remote synchronization in the star configuration with the tonic neuron at the hub. While the present study focuses on the Huber–Braun model in four-neuron motifs, excitability and electrical coupling are features shared by many conductance-based neuronal models. This motivates the hypothesis that analogous structure–dynamics relationships may also arise in other neuronal models and small network motifs, highlighting the potential role of local structural control in shaping collective neural rhythms.
Weight initialization determines the starting configuration of a neural network and can influence its subsequent optimization. In Transformer-based forecasting models, different parameter groups also have distinct structural roles that conventional initialization schemes do not explicitly capture. This study develops a structure-aware weight initialization framework for Informer in which deterministic dynamical sequences are transformed into initial weights according to the structural roles of different parameter groups. The proposed procedure combines sequence processing with tensor-specific normalization, clipping, fan-aware scaling, and module-dependent gains to control initialization on feed-forward, convolutional, and projection weights. We evaluate the framework on ETTh1, ETTh2, ETTm1, and ETTm2 benchmark datasets. Compared with the Standard Informer, the proposed configuration achieves lower mean MSE and MAE across the evaluated datasets and forecasting horizons, with the largest reductions observed on ETTh2 at horizon 48, reaching 20.73
Bilinguals rely on language control mechanisms to manage cross-language interference. Interactions between the pre-supplementary motor area (pre-SMA) and the left cerebellum play an important role in bilingual language control. However, our previous brain stimulation study targeting the cerebellum alone failed to detect significant behavioral changes in bilingual language control. In this study, we combined dual-site tDCS with fMRI to investigate its effects on bilingual language control and compared it with conventional single-site tDCS. Using a within-subject design, Chinese-English bilinguals underwent three sessions: pre-stimulation baseline, single-site tDCS over pre-SMA, and dual-site tDCS over pre-SMA and left cerebellum. Following dual-site stimulation, we observed reduced brain activity within the language control network, including the right cerebellar lobule VI, and strengthened thalamo-cerebellar connections during bilingual language production. In a separate behavioral experiment, language switch cost was lower after dual-site stimulation than at baseline but did not differ significantly between dual-site and single-site stimulation. Together, these findings provide preliminary evidence that dual-site tDCS modulates bilingual language control, potentially by strengthening thalamo-cerebellar connectivity and modulating cerebellar neural activity. These results also highlight the functional interaction between the pre-SMA and the cerebellum in supporting bilingual language control.
Spiking neural networks (SNNs) enable low-power brain-inspired computing via sparse binary spike transmission. Nevertheless, conventional homogeneous SNNs built exclusively upon leaky integrate-and-fire (LIF) neurons exhibit limited capacity for complex multi-scale temporal cognitive dynamics. Existing parametric-heterogeneous SNNs only tune internal hyperparameters within single-compartment LIF units and cannot capture diverse neuronal computations observed in biological circuits. Drawing inspiration from cross-neuronal dynamical heterogeneity in neural systems for multi-dimensional perception and cognition, this work proposes the Neurodynamical Heterogeneous Spiking Neural Network (NHSNN). NHSNN integrates three functionally distinct spiking neuron models: LIF, dendrite-inspired Pinsky–Rinzel (PR), and leaky resonate-and-fire (LRF), with each neuron type strategically assigned to dedicated functional modules to realise cross-neuronal heterogeneous spike encoding. We evaluate NHSNN across three representative cognitive-signal modalities: clinical electroencephalogram (EEG) recordings, rhythmic auditory music data, and textual semantic sequences. On EEG-diagnosis and music-classification benchmarks, NHSNN achieves accuracy on a comparable scale relative to mainstream full-precision artificial neural networks (ANNs), though cross-study comparisons require caution owing to differing experimental protocols; on textual semantic tasks, it obtains accuracy comparable to existing SNN alternatives. Under matched network dimensionality, our approach substantially reduces parameter overhead while preserving intrinsic spike-driven energy advantages. Ablation and controlled baseline experiments verify that complementary cross-neuronal dynamical synergy – rather than raw increase of model complexity – yields strong multi-modal representations and removes requirements for heavy auxiliary modules. By balancing biological inspiration, temporal modeling capacity and computational efficiency, NHSNN provides a lightweight neuromorphic paradigm for resource-constrained edge cognitive-computing scenarios.
