Neural Networks最新文献

Affective brain-computer interface is an important part of realizing emotional human-computer interaction. However, existing objective individual differences among subjects significantly hinder the application of electroencephalography (EEG) emotion recognition. Existing methods still lack the complete extraction of subject-invariant representations for EEG and the ability to fuse valuable information from multiple subjects to facilitate the emotion recognition of the target subject. To address the above challenges, we propose a Multi-source Selective Graph Domain Adaptation Network (MSGDAN), which can better utilize data from different source subjects and perform more robust emotion recognition on the target subject. The proposed network extracts and selects the individual information specific to each subject, where public information refers to subject-invariant components from multi-source subjects. Moreover, the graph domain adaptation network captures both functional connectivity and regional states of the brain via a dynamic graph network and then integrates graph domain adaptation to ensure the invariance of both functional connectivity and regional states. To evaluate our method, we conduct cross-subject emotion recognition experiments on the SEED, SEED-IV, and DEAP datasets. The results demonstrate that the MSGDAN has superior classification performance.

This study introduces an innovative neural network framework named spectral integrated neural networks (SINNs) to address both forward and inverse dynamic problems in three-dimensional space. In the SINNs, the spectral integration technique is utilized for temporal discretization, followed by the application of a fully connected neural network to solve the resulting partial differential equations in the spatial domain. Furthermore, the polynomial basis functions are employed to expand the unknown function, with the goal of improving the performance of SINNs in tackling inverse problems. The performance of the developed framework is evaluated through several dynamic benchmark examples encompassing linear and nonlinear heat conduction problems, linear and nonlinear wave propagation problems, inverse problem of heat conduction, and long-time heat conduction problem. The numerical results demonstrate that the SINNs can effectively and accurately solve forward and inverse problems involving heat conduction and wave propagation. Additionally, the SINNs provide precise and stable solutions for dynamic problems with extended time durations. Compared to commonly used physics-informed neural networks, the SINNs exhibit superior performance with enhanced convergence speed, computational accuracy, and efficiency.

Signed graphs have been widely applied to model real-world complex networks with positive and negative links, and signed graph embedding has become a popular topic in the field of signed graph analysis. Although various signed graph embedding methods have been proposed, most of them still suffer from the generality problem. Namely, they cannot simultaneously achieve the satisfactory performance in multiple downstream tasks. In view of this, in this paper we propose a signed embedding method named MOSGCN which exhibits two significant characteristics. Firstly, MOSGCN designs a multi-order neighborhood feature fusion strategy based on the structural balance theory, enabling it to adaptively capture local and global structure features for more informative node representations. Secondly, MOSGCN is trained by using the signed graph contrastive learning framework, which further helps it learn more discriminative and robust node representations, leading to the better generality. We select link sign prediction and community detection as the downstream tasks, and conduct extensive experiments to test the effectiveness of MOSGCN on four benchmark datasets. The results illustrate the good generality of MOSGCN and the superiority by comparing to state-of-the-art methods.

Inspired by the brain's information processing using binary spikes, spiking neural networks (SNNs) offer significant reductions in energy consumption and are more adept at incorporating multi-scale biological characteristics. In SNNs, spiking neurons serve as the fundamental information processing units. However, in most models, these neurons are typically simplified, focusing primarily on the leaky integrate-and-fire (LIF) point neuron model while neglecting the structural properties of biological neurons. This simplification hampers the computational and learning capabilities of SNNs. In this paper, we propose a brain-inspired deep distributional reinforcement learning algorithm based on SNNs, which integrates a bio-inspired multi-compartment neuron (MCN) model with a population coding approach. The proposed MCN model simulates the structure and function of apical dendritic, basal dendritic, and somatic compartments, achieving computational power comparable to that of biological neurons. Additionally, we introduce an implicit fractional embedding method based on population coding of spiking neurons. We evaluated our model on Atari games, and the experimental results demonstrate that it surpasses the vanilla FQF model, which utilizes traditional artificial neural networks (ANNs), as well as the Spiking-FQF models that are based on ANN-to-SNN conversion methods. Ablation studies further reveal that the proposed multi-compartment neuron model and the quantile fraction implicit population spike representation significantly enhance the performance of MCS-FQF while also reducing power consumption.

In Industrial Internet of Things (IIoT) production and operation processes, a substantial amount of video data is generated, often containing sensitive personal and commercial information. This paper proposed three new multiscroll Hopfield neural network (MHNN) systems by utilizing an improved segmented nonlinear non-ideal magnetic-controlled memristor model for electromagnetic radiation. Through dynamical methods, the constructed neural network's multidimensional multiscroll attractors and initial offset boosting behavior are analyzed. The observed initial offset boosting behavior demonstrates the system has extreme multistability. Secondly, a video encryption application based on the MHNN system is implemented on the Raspberry Pi platform. This approach encrypts each frame of the extracted video image using a novel encryption algorithm through frame-by-frame encryption, achieving significant encryption results with an information entropy calculation result of 7.9973. This provides strong protection for video data generated in IIoT. Finally, the proposed MHNN system is implemented on Field-Programmable Gate Array (FPGA) digital hardware platform.

Clustering of multi-view data divides objects into groups by preserving structure of clusters in all views, requiring simultaneously takes into consideration diversity and consistency of various views, corresponding to the shared and specific components of various views. Current algorithms fail to fully characterize and balance diversity and consistency of various views, resulting in the undesirable performance. Here, a novel Multi-View Clustering with Deep non-negative matrix factorization and Multi-Level Representation (MVC-DMLR) learning is proposed, which integrates feature learning, multi-level topology representation, and clustering of multi-view data. Specifically, MVC-DMLR first learns multi-level representation (also called deep features) of objects with deep nonnegative matrix factorization (DNMF), facilitating the exploitation of hierarchical structure of multi-view data. Then, it learns multi-level graphs for each view from multi-level representation, where relations between diversity and consistency are addressed at various resolutions. MVC-DMLR integrates multi-level representation learning, multi-level topology representation learning and clustering, which is formulated as an optimization problem. Experimental results show the superiority of MVC-DMLR to baselines in terms of accuracy, F1-score, normalized mutual information and adjusted rand index.

Sparse Bayesian learning has promoted many effective frameworks of brain activity decoding for the brain-computer interface, including the direct reconstruction of muscle activity using brain recordings. However, existing sparse Bayesian learning algorithms mainly use Gaussian distribution as error assumption in the reconstruction task, which is not necessarily the truth in the real-world application. On the other hand, brain recording is known to be highly noisy and contains many non-Gaussian noises, which could lead to large performance degradation for sparse Bayesian learning algorithms. The goal of this paper is to propose a novel robust implementation of sparse Bayesian learning so that robustness and sparseness can be realized simultaneously. Motivated by the exceptional robustness of maximum correntropy criterion (MCC), we proposed integrating MCC to the sparse Bayesian learning regime. To be specific, we derived the explicit error assumption inherent in the MCC, and then leveraged it for the likelihood function. Meanwhile, we utilized the automatic relevance determination technique as the sparse prior distribution. To fully evaluate the proposed method, a synthetic example and a real-world muscle activity reconstruction task with two different brain modalities were leveraged. Experimental results showed, our proposed sparse Bayesian correntropy learning framework significantly improves the robustness for the noisy regression tasks. Our proposed algorithm could realize higher correlation coefficients and lower root mean squared errors for the real-world muscle activity reconstruction scenario. Sparse Bayesian correntropy learning provides a powerful approach for brain activity decoding which will promote the development of brain-computer interface technology.

Generalized Category Discovery (GCD) addresses a more realistic and challenging setting in semi-supervised visual recognition, where unlabeled data contains samples from both known and novel categories. Recently, prototypical classifier has shown prominent performance on this issue, with the Softmax-based Cross-Entropy loss (SCE) commonly employed to optimize the distance between a sample and prototypes. However, the inherent non-bijectiveness of SCE prevents it from resolving intraclass relations among samples, resulting in semantic ambiguity. To mitigate this issue, we propose Distribution Consistency Regularization (DCR) for the prototypical classifier. By leveraging a simple intraclass consistency loss, we enforce the classifier to yield consistent distributions for samples belonging to the same class. In doing so, we equip the classifier to better capture local structures and alleviate semantic ambiguity. Additionally, we propose using partial labels, rather than hard pseudo labels, to explore potential positive pairs in unlabeled data, thereby reducing the risk of introducing noisy supervisory signals. DCR requires no external sophisticated module, rendering the enhanced model concise and efficient. Extensive experiments validate consistent performance benefits of DCR while achieving competitive or better performance on six benchmarks. Hence, our method can serve as a strong baseline for GCD. Our code is available at: https://github.com/yichenwang231/DCR.