Although federated learning stands as a prominent distributed learning paradigm across multiple clients, it has grappled with a dilemma: the choice between developing a singular global model to promote global generalization or nurturing personalized models to accommodate personalization. In this article, a federated learning with warm-start (WarmFed) is proposed to obtain both robust global and personalized models. First, a client knowledge-driven initialization approach named warm-start is introduced, which provides resilient global information to assist subsequent training. Then, a dynamic self-distillation (DSD) strategy is proposed to obtain more resilient personalized models and a model fine-tuning (MFT) mechanism based on synthetic data is designed to improve performance of the global model on the server side. The collaborative deployment of DSD and MFT exploits potential to achieve both generalization and personalization under warm-start. Comprehensive experiments underscore the superiority of our approach across both global and personalized models under domain shift and label skew.

Deep hashing retrieval has gained widespread use in big data retrieval due to its robust feature extraction and efficient hashing process. However, training advanced deep hashing models has become more expensive due to complex optimizations and large datasets. Coreset selection and Dataset Condensation lower overall training costs by reducing the volume of training data without significantly compromising model accuracy for classification task. In this paper, we explore the effect of mainstream dataset condensation methods for deep hashing retrieval and propose IEM (Information-intensive feature-Embedding Matching), which is centered on distribution matching and incorporates model and data augmentation techniques to further enhance the feature of hashing space. Extensive experiments demonstrate the superior performance and efficiency of our approach.

Recent work has shown the effectiveness of the plug-and-play priors (PnP) framework for regularized image reconstruction. However, the performance of PnP depends on the quality of the denoisers used as priors. In this letter, we design a novel PnP denoising prior, called multiple self-similarity net (MSSN), based on the recurrent neural network (RNN) with self-similarity matching using multi-head attention mechanism. Unlike traditional neural net denoisers, MSSN exploits different types of relationships among non-local and repeating features to remove the noise in the input image. We numerically evaluate the performance of MSSN as a module within PnP for solving magnetic resonance (MR) image reconstruction. Experimental results show the stable convergence and excellent performance of MSSN for reconstructing images from highly compressive Fourier measurements.

This paper discusses the problem of adaptive exponential synchronization in mean square for a new neural network model with the following features: 1) the noise is characterized by the Lévy process and the parameters of the model change in line with the Markovian process; 2) the master system is also disturbed by the same Lévy noise; and 3) there are multiple slave systems, and the state matrix of each slave system is an affine function of the state matrices of all slave systems. Based on the Lyapunov functional theory, the generalized Itô's formula, -matrix method, and the adaptive control technique, some criteria are established to ensure the adaptive exponential synchronization in the mean square of the master system and each slave system. Moreover, the update law of the control gain and the dynamic variation of the parameters of the slave systems are provided. Finally, the effectiveness of the synchronization criteria proposed in this paper is verified by a practical example.

Sequence information processing, for instance, the sequence memory, plays an important role on many functions of brain. In the workings of the human brain, the steady-state period is alterable. However, in the existing sequence memory models using heteroassociations, the steady-state period cannot be changed in the sequence recall. In this work, a novel neural network model for sequence memory with controllable steady-state period based on coherent spininteraction is proposed. In the proposed model, neurons fire collectively in a phase-coherent manner, which lets a neuron group respond differently to different patterns and also lets different neuron groups respond differently to one pattern. The simulation results demonstrating the performance of the sequence memory are presented. By introducing a new coherent spin-interaction sequence memory model, the steady-state period can be controlled by dimension parameters and the overlap between the input pattern and the stored patterns. The sequence storage capacity is enlarged by coherent spin interaction compared with the existing sequence memory models. Furthermore, the sequence storage capacity has an exponential relationship to the dimension of the neural network.

Excitement and synchronization of electrically and chemically coupled Newman-Watts (NW) small-world neuronal networks with a short-term synaptic plasticity described by a modified Oja learning rule are investigated. For each type of neuronal network, the variation properties of synaptic weights are examined first. Then the effects of the learning rate, the coupling strength and the shortcut-adding probability on excitement and synchronization of the neuronal network are studied. It is shown that the synaptic learning suppresses the over-excitement, helps synchronization for the electrically coupled network but impairs synchronization for the chemically coupled one. Both the introduction of shortcuts and the increase of the coupling strength improve synchronization and they are helpful in increasing the excitement for the chemically coupled network, but have little effect on the excitement of the electrically coupled one.