Fingerprint-based localization has gained significant commercial value in recent years and have brought new opportunities to location-based services in indoor environments. Existing fingerprint indoor positioning studies have focused on various neural networks (e.g., Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) and hybrid approaches). However, these methods are limited by their inherent defects, such as local trapping in CNNs and gradient extinction in RNNs, which result in low localization accuracy and high training overhead. In this paper, we devise a deep learning localization method, called Localization Transformer (LoT), to further enhance the indoor localization performance. Then, to address the problem of information loss caused by the input of the Channel State Information (CSI) matrix into LoT, we propose a Trend-Cyclical Decomposition Padding method that performs a trend-cyclical decomposition of the raw data to pad the dimensions. In addition, a Rectangular Patching strategy is designed to “patch” any shape of CSI matrix, so that the CSI matrix can be fed into a self-attention based learning module. Finally, by using a fully connected network (FCNN) to map the learned features, the 3-D coordinates can be estimated accurately. Experiments were conducted in indoor and urban canyon scenarios with three signal-to-noise ratios (SNR). The experimental results indicate the mean absolute errors of LoT in indoor and urban scenes is 0.18m and 0.48m respectively, which shows that it outperforms the CSI fingerprint-based methods regarding localization accuracy.

Multi-access edge computing (MEC) has been booming in recent years, as it promises to fulfill the growing low-latency requirements of applications on large amounts of Internet of things (IoT) devices. Nevertheless, as latency-sensitive applications tend to become more complicated, existing schemes are too sophisticated, which may result in exceeding the real-time requirements of IoT systems. In this paper, we investigate the task offloading problem for the multi-device multi-edge server IoT system integrated with MEC. Firstly, according to the performance gains obtained by offloading different subtasks, we formalize a system latency minimization problem with energy utilization consideration, which has been proven to be NP-hard. Then, to address it, we propose a heuristic computation offloading scheduling scheme, which offloads appropriate subtasks to edge servers such that the system latency is minimized. Additionally, we theoretically prove the upper and lower boundaries of the system latency. Extensive simulation results corroborate that the proposed algorithm is low-complexity yet effective in decreasing the system latency by 16.15% (and up to 51.18%), improving the energy efficiency of local devices by 8.97% (and up to 21.69%) and shortening the offloading strategy execution time by 96.33% (and up to 98.95%).

Deep learning (DL) has revolutionized time series forecasting (TSF), surpassing traditional statistical methods (e.g., ARIMA) and machine learning techniques in modeling complex nonlinear dynamics and long-term dependencies prevalent in real-world temporal data. This comprehensive survey reviews state-of-the-art DL architectures for TSF, focusing on four core paradigms: (1) Convolutional Neural Networks (CNNs), adept at extracting localized temporal features; (2) Recurrent Neural Networks (RNNs) and their advanced variants (LSTM, GRU), designed for sequential dependency modeling; (3) Graph Neural Networks (GNNs), specialized for forecasting structured relational data with spatial-temporal dependencies; and (4) Transformer-based models, leveraging self-attention mechanisms to capture global temporal patterns efficiently. We provide a rigorous analysis of the theoretical underpinnings, recent algorithmic advancements (e.g., TCNs, attention mechanisms, hybrid architectures), and practical applications of each framework, supported by extensive benchmark datasets (e.g., ETT, traffic flow, financial indicators) and standardized evaluation metrics (MAE, MSE, RMSE). Critical challenges, including handling irregular sampling intervals, integrating domain knowledge for robustness, and managing computational complexity, are thoroughly discussed. Emerging research directions highlighted include diffusion models for uncertainty quantification, hybrid pipelines combining classical statistical and DL techniques for enhanced interpretability, quantile regression with Transformers for risk-aware forecasting, and optimizations for real-time deployment. This work serves as an essential reference, consolidating methodological innovations, empirical resources, and future trends to bridge the gap between theoretical research and practical implementation needs for researchers and practitioners in the field.