Time series prediction is widely used in traffic planning, weather forecasting, and energy management. However, real-world time series often exhibit complex dynamic characteristics, such as multi-scale patterns and dynamic dependencies, making predictions challenging. Existing methods can model these patterns but struggle with component interactions and timestamp utilization. To address these issues, we propose FDS-Net (Frequency-Decomposed Synergistic Network), which enhances interactions and timestamp utilization through three core modules: Frequency-Decomposed and Reconstructed Block, which decomposes time series into components for explicit modeling and facilitates interactions through signal reconstruction; Multi-Scale Temporal Semantic Extraction Block, which uses multi-scale convolutions to extract diverse temporal features; and Temporal Feature Interaction Block which integrates timestamp features with reconstructed signals and uses attention mechanisms to capture dynamic dependencies. Additionally, we introduce an adaptive dynamic weighted loss function to optimize training. Experimental results show FDS-Net’s superior performance in long-term and short-term forecasting, providing new insights for time series prediction.

Multivariate time series prediction is widely applied across various fields. Recently, MLP-based forecasting methods have shown potential due to their efficiency. However, most of these methods analyze complex time series data primarily from a global perspective in the time domain, leading to incomplete feature extraction and affecting prediction accuracy. To address this issue, we propose TF-DPNet, a lightweight Time Frequency Integrated Network with Dual Perspectives for multivariate time series prediction. TF-DPNet leverages both frequency and time domain information, treating the global and local perspectives of time series data as separate inputs to enhance predictive capability. Specifically, TF-DPNet processes these inputs by extending frequency domain information and reconstructing time domain features to capture rich characteristics. Subsequently, a time feature extractor further enhances time domain feature extraction. Finally, the features extracted from dual perspectives are integrated and optimized using a feature amplifier to improve prediction performance. TF-DPNet’s performance is validated on real-world datasets, showing superior forecasting accuracy and computational efficiency.

Autonomous aerial vehicles (AAVs) have found widespread utility in the field of multi-target tracking (MTT) due to their inherent advantages, such as ease of deployment, flexible maneuverability, and cooperative communication capabilities. AAVs can perform tasks ranging from regional surveillance to tracking and search-and-rescue operations in hazardous environments. Nonetheless, the issue of how to efficiently coordinate multiple AAVs to track diverse mobile targets remains a critical concern. This paper focuses on MTT with multi-AAV, aiming at optimizing system performance in scenarios where network availability is limited. In contrast to treating sensor perception as a monolithic process, we propose a scheme for cooperative sensing and data processing. This scheme is designed to reduce system response latency in environmental information sensing and multidimensional data processing for multiple AAVs. Furthermore, in contrast to assuming linear target movement and single-step target position prediction, we introduce a multi-agent deep reinforcement learning (MADRL) framework combined with multi-step prediction extended Kalman Filter (MP-EKF). This framework is tailored to enhance tracking precision, especially when targets’ trajectories are curved, which can reduce AAV flight displacement if the target’s position can be predicted multiple steps later. In addition, unlike using latency as a real-time application to measure the “freshness” of information, the Age of Information (AoI) is introduced for considering the waiting time of transmission and calculation between multiple AAVs. This assessment method is utilized to comprehensively evaluate and mitigate data latency within the MADRL algorithm. Finally, extensive simulation experiments demonstrate that the proposed scheme significantly outperforms both baseline methods and state-of-the-art approaches in terms of AoI, system latency, and energy consumption.

Multi-Unmanned Aerial Vehicle (UAV)-assisted Mobile Edge Computing (MEC) technology can provide users with more flexible and efficient computing services. Most of the existing research works realize multi-UAV collaboration through single-hop or multi-hop task relaying. Although the delay and energy consumption incurred in transmitting mission results to the destination node is relatively small, the UAV node changes dynamically and the result feedback should not be ignored. At the same time, when the remaining power of the UAVs swarms is unevenly distributed, when one UAV runs out of power, the users in the area it serves will fall into congestion again. In order to solve these problems, we propose an energy-aware computation offloading and routing (ECOR) strategy with result feedback. ECOR allows tasks to select the best node by flying ad-hoc network (FANET) multi-hop routing, thus increasing the lifetime of the entire network and the number of tasks completed during the lifetime, i.e., the lifetime value. Meanwhile, we propose to embed the remaining energy and reliability information of the UAV in the Hello message, thus making the UAV energy-aware. Considering the dynamics of the environment and the complexity of the optimization objective, we propose to use Dueling DQN to optimize offloading and routing decisions. After the task is successfully offloaded, the system will enter the result transmission phase to relay the task result to the destination node. Simulation results show that ECOR significantly outperforms other strategies in terms of cumulative rewards, energy efficiency, and task delivery rates.

Identifying heterogeneous cognitive impairment markers at an early stage is vital for Alzheimer's disease diagnosis. However, due to complex and uncertain brain connectivity features in the cognitive domains, it remains challenging to quantify functional brain connectomic changes during non-pharmacological interventions for amnestic mild cognitive impairment (aMCI) patients. We present a quantitative method for functional brain network analysis of fMRI data based on the multi-graph unsupervised Gaussian embedding method (MG2G). This neural network-based model can effectively learn low-dimensional Gaussian distributions from the original high-dimensional sparse functional brain networks, quantify uncertainties in link prediction, and discover the intrinsic dimensionality of brain networks. Using the Wasserstein distance to measure probabilistic changes, we discovered that brain regions in the default mode network and somatosensory/somatomotor hand, fronto-parietal task control, memory retrieval, and visual and dorsal attention systems had relatively large variations during non-pharmacological training, which might provide distinct biomarkers for fine-grained monitoring of aMCI cognitive alteration. An important finding of our study is the ability of the new method to capture subtle changes for individual patients before and after short-term intervention. More broadly, the MG2G method can be used in studying multiple brain disorders and injuries, e.g., in Parkinson's disease or traumatic brain injury (TBI), and hence it will be useful to the wider neuroscience community.

Spontaneous brain activity studies have revealed `small-world' property in functional networks based on correlated or positively-correlated relationships. However, studies neither investigated negatively-correlated functional networks, nor checked the `dynamic' properties of the whole functional organization. After subjects performed a specific task, what changes will be caused in the intrinsic organization? In this study, we examined pre-task and post-task resting brains using functional MRI (fMRI). Then we suggested two perspectives with positively-correlated brain functional network (PCBFN) and negatively-correlated brain functional network (NCBFN), by testing whether correlation coefficients were positive or negative. The major findings were: 1) the PCBFN was with small-world architecture but the NCBFN was not, with different motifs, and the PCBFN showed stronger small-world effect at the post-task resting state; 2) Both PCBFN and NCBFN followed an exponentially truncated power law degree distribution. This study may offer a framework to investigate the intrinsic organization of spontaneous brain activity.