Neural Networks最新文献

In neuroscience, phase synchronization (PS) is a crucial mechanism that facilitates information processing and transmission between different brain regions. Specifically, global phase synchronization (GPS) characterizes the degree of PS among multivariate neural signals. In recent years, several GPS methods have been proposed. However, they primarily focus on the collective synchronization behavior of multivariate neural signals, while neglecting the structural difference between oscillator networks. Therefore, in this paper, we introduce a method named total correlation-based synchronization (TCS) to quantify GPS intensity by examining network organization. To evaluate the performance of TCS, we conducted simulations using the Rössler model and compared it to three existing methods: circular omega complexity, hyper-torus synchrony, and symbolic phase difference and permutation entropy. The results indicate that TCS outperforms the other methods at distinguishing the GPS intensity between networks with similar structures. And it offers insight into the separation and integration behavior of signals during synchronization. Furthermore, to validate this method with experimental data, TCS was applied to analyze the GPS variation of multichannel stereo-electroencephalography (SEEG) signals recorded from onset zones of patients with temporal lobe epilepsy. It was observed that the termination of seizures was associated with the increased GPS and the integration of brain regions. Taken together, TCS offers an alternative way to measure GPS of multivariate signals, which may shed new lights on the mechanism of brain functions and neurological disorders, such as learning, memory, epilepsy, and Alzheimer's disease.

There is currently strong interest in improving visual object tracking by augmenting the RGB modality with the output of a visual event camera that is particularly informative about the scene motion. However, existing approaches perform event feature extraction for RGB-E tracking using traditional appearance models, which have been optimised for RGB only tracking, without adapting it for the intrinsic characteristics of the event data. To address this problem, we propose an Event backbone (Pooler), designed to obtain a high-quality feature representation that is cognisant of the innate characteristics of the event data, namely its sparsity. In particular, Multi-Scale Pooling is introduced to capture all the motion feature trends within event data through the utilisation of diverse pooling kernel sizes. The association between the derived RGB and event representations is established by an innovative module performing adaptive Mutually Guided Fusion (MGF). Extensive experimental results show that our method significantly outperforms state-of-the-art trackers on two widely used RGB-E tracking datasets, including VisEvent and COESOT, where the precision and success rates on COESOT are improved by 4.9% and 5.2%, respectively. Our code will be available at https://github.com/SSSpc333/TENet.

This article proposes a novel deep clustering model based on the variational autoencoder (VAE), named GamMM-VAE, which can learn latent representations of training data for clustering in an unsupervised manner. Most existing VAE-based deep clustering methods use the Gaussian mixture model (GMM) as a prior on the latent space. We employ a more flexible asymmetric Gamma mixture model to achieve higher quality embeddings of the data latent space. Second, since the Gamma is defined for strictly positive variables, in order to exploit the reparameterization trick of VAE, we propose a transformation method from Gaussian distribution to Gamma distribution. This method can also be considered a Gamma distribution reparameterization trick, allows gradients to be backpropagated through the sampling process in the VAE. Finally, we derive the evidence lower bound (ELBO) based on the Gamma mixture model in an effective way for the stochastic gradient variational Bayesian (SGVB) estimator to optimize the proposed model. ELBO, a variational inference objective, ensures the maximization of the approximation of the posterior distribution, while SGVB is a method used to perform efficient inference and learning in VAEs. We validate the effectiveness of our model through quantitative comparisons with other state-of-the-art deep clustering models on six benchmark datasets. Moreover, due to the generative nature of VAEs, the proposed model can generate highly realistic samples of specific classes without supervised information.

Node classification is a fundamental task of Graph Neural Networks (GNNs). However, GNN models tend to suffer from the class imbalance problem which deteriorates the representation ability of minority classes, thus leading to unappealing classification performance. The most straightforward and effective solution is to augment the minority samples for balancing the representations of majority and minority classes. Previous methods leverage a limited number of labeled nodes to generate new samples, without considering the overall class characteristics and failing to reflect the underlying class distributions. Besides, they often yield less distinguishable nodes that cannot represent their original classes well, because they may incorporate useless information from other classes to form node representations. To address this issue, we propose a Class Distribution-aware Conditional Generative Adversarial Network (CDCGAN) to generate diverse and distinguishable minority nodes based on their class distribution characteristics. Specifically, we extract the node embeddings and class distributions while preserving the topology and attribute information, thus capturing the overall class characteristics. Then, the obtained class distributions are used to design a conditional generator, which incorporates nonlinear transformations to generate diverse minority nodes and leverages adversarial learning to maintain intrinsic class distribution characteristics. At last, to ensure the distinguishability of node representations, a unique discriminator is implemented to jointly discriminate and classify nodes of the augmented graph. Extensive experiments conducted on six datasets demonstrate that the proposed CDCGAN outperforms the state-of-the-art methods on widely used evaluation metrics. The source code is available at https://github.com/Crystal-LiuBojia/CDCGAN.

This paper studies the asynchronous output feedback control and H_∞ synchronization problems for a class of continuous-time stochastic hidden semi-Markov jump neural networks (SMJNNs) affected by actuator saturation. Initially, a novel neural networks (NNs) model is constructed, incorporating semi-Markov process (SMP), hidden information, and Brownian motion to accurately simulate the complexity and uncertainty of real-world environments. Secondly, acknowledging system mode mismatches and the need for robust anti-interference capabilities, a non-fragile controller based on hidden information is proposed. The designed controller effectively mitigates the impact of uncertainties enhancing system reliability. Furthermore, sufficient conditions for stochastic mean square synchronization (MSS) within the domain of attraction are provided, and optimal control is achieved through the construction of a Lyapunov function based on SMP. Finally, the feasibility of the proposed method is verified through numerical examples.

Standard makeup transfer techniques mainly focus on facial makeup. The texture details of headwear in style examples tend to be ignored. When dealing with complex portrait style transfer, simultaneous correct headwear and facial makeup transfer often cannot be guaranteed. In this paper, we construct the Peking Opera makeup dataset and propose a makeup transfer network for Opera faces called OperaGAN. This network consists of two key components: the Makeup and Headwear Style Encoder module (MHSEnc) and the Identity Coding and Makeup Fusion module (ICMF). MHSEnc is specifically designed to extract the style features from global and local perspectives. ICMF extracts the source image's facial features and combines them with the style features to generate the final transfer result. In addition, multiple overlapping local discriminators are utilized to transfer the high-frequency details in opera makeup. Experiments demonstrate that our method achieves state-of-the-art results in simultaneously transferring opera makeup and headwear. And the method can transfer headwear with missing content and controllable intensity makeup. The code and dataset will be available at https://github.com/Ivychun/OperaGAN.

Multi-hop path completion is a key part of temporal knowledge graph completion, which aims to infer complex relationships and obtain interpretable completion results. However, the traditional multi-hop path completion models mainly focus on the static knowledge graph with sufficient relationship instances, and does not consider the impact of timestamp information on the completion path, and is not suitable for few-shot relations. These limitations make the performance of these models not good when dealing with few-shot relationships in temporal knowledge graphs. In order to issue these challenges, we propose the Few-shot Temporal knowledge graph completion model based on the Multi-hop Interpretable meta-learning(FTMI). First, by aggregating the multi-hop neighbor information of the task relationship to generate a time-aware entity representation to enhance the task entity representation, the introduction of the timestamp information dimension enables the FTMI model to understand and deal with the impact of time changes on entities and relationships. In addition, time-aware entity pair representations are encoded using Transformer. At the same time, the specific representation of task relationship is generated by means of mean pooling layer aggregation. In addition, the model applies the reinforcement learning framework to the whole process of multi-hop path completion, constructs the strategy network, designs the new reward function to achieve the balance between path novelty and length, and helps Agent find the optimal path, thus realizing the completion of the temporal knowledge graph with few samples. In the training process, meta-learning is used to enable the model to quickly adapt to new tasks in the case of few samples. A huge number of experiments were carried out on two datasets to validate the model's validity.

Social media platforms, rich in user-generated content, offer a unique perspective on public opinion, making stance detection an essential task in opinion mining. However, traditional deep neural networks for stance detection often suffer from limitations, including the requirement for large amounts of labeled data, uninterpretability of prediction results, and difficulty in incorporating human intentions and domain knowledge. This paper introduces the First-Order Logic Aggregated Reasoning framework (FOLAR), an innovative approach that integrates first-order logic (FOL) with large language models (LLMs) to enhance the interpretability and efficacy of stance detection. FOLAR comprises three key components: a Knowledge Elicitation module that generates FOL rules using a chain-of-thought prompting method, a Logic Tensor Network (LTN) that encodes these rules for stance detection, and a Multi-Decision Fusion mechanism that aggregates LTNs' outputs to minimize biases and improve robustness. Our experiments on standard benchmarks demonstrate the effectiveness of FOLAR, showing it as a promising solution for explainable and accurate stance detection. The source code will be made publicly available to foster further research.

Memristors are commonly used as the connecting parts of neurons in brain-like neural networks. The memristors, unlike the existing literature, possess the capability to function as both self-connected synaptic weights and interconnected synaptic weights, thereby enabling the generation of intricate initials-regulated plane coexistence behaviors. To demonstrate this dynamical effect, a Hopfield neural network with two-memristor-interconnected neurons (TMIN-HNN) is proposed. On this basis, the stability distribution of the equilibrium points is analyzed, the related bifurcation behaviors are studied by utilizing some numerical simulation methods, and the plane coexistence behaviors are proved theoretically and revealed numerically. The results clarify that TMIN-HNN not only exhibits complex bifurcation behaviors, but also has initials-regulated plane coexistence behaviors. In particular, the coexistence attractors can be switched to different plane locations by the initial states of the two memristors. Finally, a digital experiment device is developed based on STM32 hardware board to verify the initials-regulated plane coexistence attractors.

The fractional-order gradient descent (FOGD) method has been employed by numerous scholars in Artificial Neural Networks (ANN), with its superior performance validated both theoretically and experimentally. However, current FOGD methods only apply fractional-order differentiation to the loss function. The application of FOGD based on Autograd to hidden layers leverages the characteristics of fractional-order differentiation, significantly enhancing its flexibility. Moreover, the implementation of FOGD in the hidden layers serves as a necessary foundation for establishing a family of fractional-order deep learning optimizers, facilitating the widespread application of FOGD in deep learning. This paper proposes an improved fractional-order gradient descent (IFOGD) method based on Multilayer Perceptron (MLP). Firstly, a fractional matrix differentiation algorithm and its fractional matrix differentiation solver is proposed based on MLP, ensuring that IFOGD can be applied within the hidden layers. Subsequently, we overcome the issue of incorrect backpropagation direction caused by the absolute value symbol, ensuring that the IFOGD method does not cause divergence in the value of the loss function. Thirdly, fractional-order Autograd (FOAutograd) is proposed based on PyTorch by reconstructing Linear layer and Mean Squared Error Loss module. By combining FOAutograd with first-order adaptive deep learning optimizers, parameter matrices in each layer of ANN can be updated using fractional-order gradients. Finally, we compare and analyze the performance of IFOGD with other methods in simulation experiments and time series prediction tasks. The experimental results demonstrate that the IFOGD method exhibits performances.