Publications

With the widespread adoption of recommendation systems, efficiently utilizing data distributed across different communities or edge devices has become a core challenge. Traditional centralized learning methods struggle to meet the demands of data privacy and transmission constraints, while Federated Learning (FL) offers a viable solution for distributed collaborative learning. However, existing FL methods (e.g., FedGCN) face limitations in handling data heterogeneity and achieving personalized models. These challenges are particularly pronounced in non-independent and identically distributed (non-IID) data scenarios, where knowledge collapse is prone to occur. This study proposes a personalized federated learning method based on FedGCN to optimize recommendation systems. By calculating the similarity of embedding vectors between clients, we dynamically adjust aggregation weights during the model aggregation process, thereby generating personalized models that adapt to heterogeneous data distributions. Experimental results demonstrate that, compared to existing methods, our approach achieves significant improvements in both convergence speed and prediction accuracy, particularly excelling in scenarios with significant data heterogeneity across communities.

One type of degenerate (or, singular) Hopf bifurcations determine the appearance of multiple limit cycles under system parameter perturbations. In the study of these degenerate Hopf bifurcations, computational formulas for the stability indexes (i.e., curvature coefficients) are essential. However, such formulas are very difficult to obtain, and hence are usually obtained by different approximation techniques. In this paper, by using engineering feedback system methodology and harmonic balance approximation techniques, these formulas are derived in the frequency domain. The results obtained in this paper can be used to study a nonlinear system within regions of one periodic solution, instead of directly dealing with multiple limit cycles. Thus, the complex multiplicity of nonlinear system dynamical behavior can be avoided.

Many real-world complex networks display a small-world feature-a high degree of clustering and a small average distance. We show that the maximum synchronizability of a network is completely determined by its associated feedback system, which has a precise meaning in terms of synchronous communication. We introduce a new concept of synchronizability matrix to characterize the maximum synchronizability of a network. Several new concepts, such as sensitive edge and robust edge, are proposed for analyzing the robustness and fragility of synchronization of a network. Using the knowledge of synchronizability, we can purposefully increase the robustness of the network synchronization and prevent it from attacks. Some applications in small-world networks are also discussed briefly.

This paper studies robust adaptive flocking control of multi-agent systems with nonlinear dynamics. In this setting, the coupling weights, perturbed by asymmetric uncertain parameters, are dynamically updated while the network topology for velocity is fixed. By designing the coupling weights according to the distributed local information on the linked nodes, flock formation can be reached if the network is connected. It is found that flock formation can be achieved if the topology of the updated coupling weights forms a spanning tree. A simulation example is given to illustrate the theoretical analysis.

A memristor with coexisting pinched hysteresis loops and twin local activity domains is presented and analyzed, with an emulator being designed and applied to the classic Chua’s circuit to replace the diode. The memristive system is modeled with four coupled first-order autonomous differential equations, which has three equilibria determined by three static equilibria of the memristor but not controlled by the system parameters. The complex dynamics of the system are analyzed by using compound coexisting bifurcation diagrams, Lyapunov exponent spectra and phase portraits, including point attractors, limit cycles, symmetrical chaotic attractors and their blasting, extreme multistability, state-switching without parameter, and transient chaos. Of particular surprise is that the extreme multistability of the system is hidden and symmetrically distributed. It is found that the existence of transient chaos in the specified parameter domain is determined by using bifurcation diagrams within different time durations and Lyapunov exponents with chaotic sequences. Finally, the symmetrical chaotic attractor and the system blasting are verified by digital signal processing experiments, which are consistent with the numerical analysis.

No abstract is provided for this article.

We investigate the constrained optimization of excitatory synaptic input patterns to fastest generate given number of spikes in theta neuron model. Optimal input timings and strengths are identified by using phase plane arguments for discrete input kicks with a given total magnitude. Furthermore, analytical results are conducted to estimate the firing time of given number of spikes resulting from a given input train. We obtain the fastest strategy as the total input size increases. In particular, when the parameter − b is large and total input size G is not so large, there are two candidate strategies to fastest achieve given number of spikes, which depend on the considered parameters. The fastest strategy for some cases of G ≫ − b to fire m spikes should partition m spikes into m − n + 1 spikes for the highest band, with largest g , and one spike for each subsequent n − 1 band. When G is sufficiently large, big kick is the fastest strategy. In addition, we establish an optimal value for the dependent variable, θ , where each input should be delivered in a non-threshold-based strategy to fastest achieve given output of subsequent spikes. Moreover, we find that reset and kick strategy is the fastest when G is small and G ≫ − b . The obtained results can lead to a better understanding of how the period of nonlinear oscillators are affected by different input timings and strengths.

It has been discovered recently that many social, biological and ecological systems have the so-called small-world and scale-free features, which has provoked new research interest in the studies of various complex networks. Yet, most network models studied thus far are binary, with the linking strengths being either 0 or 1, while which are best described by weighted-linking networks, in which the vertices interact with each other with varying strengths. Here we found that the distribution of connection strengths of scientific collaboration networks decays also in a power-law form and we conjecture that all weighted-linking networks of this type follow the same distribution.

No abstract is provided for this article.

A feedback control method is developed for controlling an unknown chaotic system, where the design is accomplished using only a finite set of one-dimensional (1-D) time-series data generated by the unknown system. The approach consists of three steps: (1) using the available time-series data to estimate a delayed time constant and an embedding dimension that both can be used to reconstruct a state space model preserving the chaotic attractor of the unknown system; (2) based on the reconstructed model, a lower order polynomial system of the same dimension is identified as a platform for controller design; and (3) a design method using the lower order approximate model is devised for determining a simple nonlinear feedback controller that works for the originally unknown underlying chaotic system. Computer simulation results on the Duffing oscillator are shown to demonstrate the effectiveness of this new feedback-control methodology.

This paper investigates the complex dynamical phenomena such as bifurcations and chaos in a structure of strongly coupled neural oscillators. Bifurcation diagrams and topological properties of these periodic solutions are then obtained, where synchronization phenomena are completely classified. A new bursting response is also observed and discussed.