The Audio-visual Classification (AVC) task aims to determine video categories by integrating audio and visual signals. Traditional methods for AVC leverage Artificial Neural Networks (ANNs) that operate on floating-point features, affording large parameter counts and consuming extensive energy. Recent research has shifted towards brain-inspired Spiking Neural Networks (SNNs), which transmit audiovisual information through sparser 0/1 spike features allowing for better energy efficiency. However, a byproduct of such sparsity is the increased difficulty in effectively encoding and utilizing these spike features. Moreover, the spike firing characteristics based on neuron membrane potential cause asynchronous spike activations due to the heterogeneous distributions of different modalities in the AVC task, resulting in cross-modal asynchronization. This issue is often overlooked by prior SNN models, resulting in lower classification accuracy compared to traditional ANN models. To address these challenges, we present a new Multimodal Interaction Spiking Network (MISNet), the first to successfully balance both accuracy and efficiency for the AVC task. As the core of MISNet, we propose a Multimodal Leaky Integrate-and-fire (MLIF) neuron, which coordinates and synchronizes the spike activations of audiovisual signals within a single neuron, distinguishing it from the prior paradigm of SNNs that relies on multiple separate processing neurons. As a result, our MISNet enables to generate audio and visual spiking features with effective cross-modal fusion. Additionally, we propose to add extra loss regularizations before fusing the obtained audio-visual features for final classification, thereby benefiting unimodal spiking learning for multimodal interaction. We evaluate our method on five audio-visual datasets, demonstrating advanced performance in both accuracy and energy consumption.

In the Internet of Things, openness significantly increases security risks. Therefore, detecting such attacks through anomaly traffic monitoring is crucial for ensuring network security. Although existing intrusion detection systems (IDSs) have made some progress in detection performance using deep learning techniques (e.g., graph learning and meta-learning), the class imbalance problem remains a major challenge that needs to be addressed. To this end, we propose a double-anchor contrastive learning (DACL) framework. First, we introduce a positive–negative sample construction module. Compared to traditional contrastive learning methods, this module directly constructs sample pairs, preserving more flow information and thereby enhancing the effectiveness. Second, we design a feature embedding and extraction module. By utilizing convolutional neural networks (CNNs) to extract spatial correlations among raw features, we generate more discriminative feature representations. Subsequently, we propose the DACL module, introducing a double-anchor mechanism to impose double constraints on interclass distances, further enhancing feature representation capabilities and clustering effects. Finally, we map the learned features to a low-dimensional space via a classification network, achieving precise detection of abnormal traffic. Extensive experiments demonstrate that DACL outperforms existing works in terms of accuracy on datasets such as CIC-DDoS2017, CIC-IDS2018, and CIC-DDoS2019.

In Cognitive Radio Networks (CRNs), network fault analysis, traffic tracing, and resource optimization are challenging tasks. With the increasing number of wireless applications and the conflict with limited wireless spectrum resources, the Sniffers Channel Assignment problem in CRNs has become particularly important. To address this issue, we propose a Value Decomposition Networks-based channel selection (CSVDN) algorithm. During centralized training, the Monitoring Quality Network (MQN) is trained based on observed data, using global information to calculate the Quality of Monitoring (QoM), which is then used as a reward to guide sniffers in selecting the optimal channels. During decentralized execution, sniffers share model parameters and independently run the MQN, sequentially selecting the optimal channels. This process ensures that sniffers collectively maximize network coverage while maintaining distributed control, thereby improving efficiency and scalability in dynamic environments. The results from NS-3 simulations show that CSVDN provides a distributed and implementable channel selection solution with high scalability and practicality, making it particularly suitable for large-scale CRNs.

Spiking Neural Networks (SNNs) have emerged as a promising alternative to traditional Artificial Neural Networks (ANNs) due to their biological interpretability. SNNs process input information in both spatial and temporal dimensions using binary spikes, offering advantages such as low energy consumption and compatibility with neural devices. Previous works have demonstrated that directly training SNNs from scratch has great potential. It can achieve performance comparable to traditional ANNs while significantly reducing energy consumption and the number of required time steps. Moreover, it outperforms ANN-SNN conversion methods. However, prior studies typically employ a fixed timestep during both training and inference, leading to two main problems: 1) During training, it results in complex gradient propagation and gradient estimation errors. 2) During inference, it causes energy waste and performance bottlenecks. To address these issues, we propose a novel strategy called Momentum Performance Queue Training and Confidence Score Inference (MPQ-CS). Firstly, MPQ-CS dynamically adjusts the timestep during the training stage based on the current training performance of the SNN using a Momentum Performance Queue (MPQ). Additionally, MPQ-CS dynamically adjusts the timestep during the inference stage using a Confidence Score (CS) based on the difficulty of input samples. Extensive experiments on multiple datasets and different SNN architectures demonstrate that MPQ-CS significantly enhances the overall performance of SNNs during both training and inference stages.

Autonomous obstacle avoidance is a critical capability for Autonomous Underwater Vehicles (AUVs) to operate safely in dynamic and uncertain marine environments. Traditional AUV control methods rely on precise physical modeling and preset rules, yet they struggle to adapt to multiple sources of uncertainty, such as random initial states, dynamic obstacles, and varying currents. In recent years, deep reinforcement learning has provided a new avenue for data-driven adaptive policy learning. However, it remains insufficient for handling long-horizon tasks with sparse rewards. While hierarchical reinforcement learning can mitigate reward sparsity through temporal abstraction, it often faces challenges including exploration–exploitation imbalance, slow global convergence, and insufficient safety guarantees. Furthermore, most existing studies neglect dynamic environmental disturbances and task continuity, which further limits the practical application of these algorithms. To address these challenges, this paper proposes a hierarchical curiosity-driven AUV obstacle avoidance algorithm (HDAO), designed for autonomous obstacle avoidance in dynamic and uncertain underwater environments. The core design of HDAO incorporates several key innovations. Firstly, it introduces a Collision Threat Index for dynamic obstacles, which enables explicit risk perception and quantifies collision threats, thereby enhancing the policy’s generalization and robustness. Secondly, a task-decoupled hierarchical architecture is employed to synergistically optimize global path planning and local obstacle avoidance behaviors. This approach effectively manages long-horizon navigation tasks while alleviating high-dimensional training pressure. Finally, a novel reward mechanism is designed by integrating hierarchical active exploration with curiosity-driven passive exploration. This mechanism effectively incentivizes the agent to explore unvisited areas under sparse reward conditions and dynamically balances exploration and exploitation. Experimental results demonstrate that HDAO significantly outperforms existing methods in terms of obstacle avoidance success rate, training convergence speed and robustness against external disturbances.