The integration of edge computing and satellite networks has emerged as a promising solution to support remote terrestrial computation with wide coverage and low latency. However, single-satellite computing leads to uneven resource utilization and degraded service quality. To address this, peer offloading is required to improve both service quality and resource efficiency. Asides from peer offloading, diverse service requests also call for an appropriate service deployment strategy, which should be jointly optimize with offloading decision. In this paper, taking into processing cost and service update cost, we formulate a long-term optimization for service deployment and peer offloading. To pursue long-term performance, the problem is first reformulated into a sequence of time-invariant problems. Since frequent service deployment adjustments incur overhead and may cause service interruption, we decompose the time-invariant problem into a service deployment subproblem and a peer offloading subproblem, optimized at different timescales. A hierarchical method iteratively solve the two subproblems. In particular, we propose an online distributed algorithm for small-timescale peer offloading. Each local peer offloading problem is transformed into a capacity-constrained minimum cost maximum flow problem, enabling a low-complexity solution via the successive shortest path algorithm. We provide theoretical analysis showing that the proposed algorithm asymptotically approaches the offline optimum at the expense of system congestion. Moreover, we show that the performance bound grows with the large-timescale interval. Simulations results validate the theoretical analysis and demonstrate the effectiveness of the propose algorithm in terms of processing cost and service update cost.

This paper aims to develop an effective uplink access scheduling strategy for massive Internet-of-Things (IoT) networks. To better reap the benefits of uplink resources, the BS has to adjust the uplink resources relies on the network states available at the BS. However, in massive IoT networks, the acquisition of network states, including traffic arrivals, channel conditions, and energy supply rate, are typically obtained through in-band feedback from devices. Therefore, the network states available at the BS are differently outdated across devices, as the staleness depends on the time elapsed since each device was last scheduled. This motivates us to develop a proactive scheduling scheme that enables the BS to schedule uplink access under differently outdated states information. To combat the performance loss caused by the outdated states information, we propose a novel primal-dual online learning framework. This framework leverages mini-batch gradient descent for dual updates and employs Online Convex Optimization for proactive primal updates, which effectively predicting current network states based on outdated knowledge. We evaluate the performance of the proposed proactive scheduling scheme against the offline optimum, which is optimized using prior knowledge of network states. The performance analysis shows that the proactive scheme asymptotically approaches to the offline optimum. Simulation results further validate the effectiveness of the proposed algorithm by comparing to other benchmarks.

Federated learning (FL) is a privacy-preserving technique that enables local model training on devices without raw data sharing. However, a critical challenge in FL lies in the communication requirement of uploading the trained models to servers, which can be hindered by interference from ambient devices, particularly in unreliable wireless environments. To address this, hierarchical FL (HFL) introduces an additional intermediate layer where the edge server performs work aggregation from the devices nearby, aiming at reducing the communication load and improving the efficiency of model training. However, existing approaches suffer from two critical limitations. First, they fail to fully quantify the impact of device competition-induced interference on transmission performance, which leads to unacceptably high upload latency and low success upload probability (SUP). Second, they lack a targeted optimization strategy to balance model accuracy and transmission efficiency under dynamic interference conditions. To address these critical limitations and mitigate their adverse impacts on FL performance, we take these gaps as the core motivation of our work and propose a targeted solution. Specifically, we first model the network as a two-layer binomial point process (BPP), which allows us to analyze the network-layer performance and calculate the SUP for the trained model. Based on this model, we propose optimizing cluster selection to balance accuracy and latency, thereby enhancing overall FL performance. We formulate this optimization as a Markov decision process (MDP) and solve it using a twin-delayed deep deterministic policy gradient (TD3)-based cluster selection algorithm (CS-TD3). In addition, to guarantee network-layer performance and enhance the efficiency of HFL, we employ an experimental exhaustive search algorithm to find the best solution within a limited range. The experimental results show that our algorithm overperforms other commonly-used algorithms in terms of HFL accuracy and model transmission latency, achieving a 10.95% improvement over the other methods.

Trajectory similarity is a fundamental operation in spatial-temporal data mining with wide-ranging applications. However, trajectories inherently exhibit diversity due to varied sampling and distribution of trajectory points, influenced by different motion patterns, sampling methods, and route constraints. This diversity leads to varying results in trajectory similarity measures, including DTW, LCSS, ED, Hausdorff, EDR, etc. In this paper, we argue for a comprehensive consideration of various distance metrics to enhance accuracy. To address this, this paper proposes a Multi-source Deep Trajectory Representation Learning method for accurate and efficient similarity queries. In particular, MDTRL comprises two key modules: (1) A novel trajectory representation module that incorporates an attention-based embedding mechanism and a deep metric learning network aggregating multiple measures. (2) A continuous metric learning strategy that adaptively updates similarity, thereby enhancing the accuracy of similarity queries. We employ the locality sensitive hashing index to further improve the similarity query. Extensive experiments conducted on real trajectory datasets reveal that MDTRL has state-of-the-art solutions, in terms of both effectiveness and efficiency across multi-source trajectories. It achieves 5x-15x speedup and 10%-15% accuracy improvement over Euclidean, Hausdorff, DTW, and Discrete Fréchet measures.

With the popularization of mobile applications and the timely acquisition of fresh data, real-time clustering and its evolution analysis have become the primary operations for data processing and knowledge discovery. Such continuous queries on massive objects are computation-intensive tasks in dynamic scenarios. However, existing clustering techniques are incompetent to achieve decent performance when computation-intensive operations frequently occur in streaming scenarios, which is caused by two challenges: (i) uncertainty of the clustering frequency; (ii) unpredictable distribution evolution. Hence, it is critical to find a lightweight model that can cluster the high-speed dynamic instances while exploiting the evolution amid different clustering results. This paper focuses on the problem of real-time clustering on streaming data in computation-intensive and high-dynamics tasks, through a framework Ocean, consisting of the Online clustering algorithm and evolution analysis. Particularly, the framework conceives a flexible composite window to augment the knowledge mining, achieving a proper real-time response in various scenarios. The evolution analysis supports full life-cycle detection, improving the adaptability to dynamic concept drifts and multiple patterns. Inspired by the grid partition strategy, this framework adopts grid feature vectors to capture the significant changes in streaming data. Furthermore, we propose an optimization that removes sparse grids timely and performs the online clustering adaptively for space and time efficiency. It is proven to be effective both theoretically and experimentally. This strategy enables real-time clustering for dynamic streaming data without degrading the clustering quality or increasing the computation cost. Experiments on real datasets and synthetic datasets verify the accuracy and effectiveness of Ocean compared to the state-of-the-art approaches, as well as the superior ability to perform clustering in a real-time manner.

Frame deletion detection is a challenging task in the field of digital video forensics. This paper proposes a deep-learning-based frame deletion detection method for single-shot videos. We capture traces of frame deletion forgery from both adjacent and long-range continuous frames. Specifically, we propose a novel multi-scale difference convolutional neural network (MSDCNN) structure, which models different levels of inter-frame variations. Then, we use the long-short-term memory network (LSTM) to capture the long-term variation pattern of multi-scale differential features. The proposed method is a simple and principled frame deletion detection framework with a small computational cost. According to the experiments, the proposed framework can achieve a more advanced performance of frame deletion detection than traditional methods and methods based on 3D convolutions.

With the prevalence of location-based services (LBS), trajectories are being generated rapidly. As is widely used in LBS, top-k trajectory similarity query serves as a key operation, deeply empowering applications such as travel route recommendation and carpooling. Given the rise of deep learning, trajectory representation has been well-proven to speed up this operator. However, existing representation-based computing modes remain two major problems understudied: the low quality of trajectory representation and insufficient support for various trajectory similarity metrics, which make them difficult to apply in practice. Therefore, we propose an Accurate metric-based representation learning approach for fast top-k trajectory similarity query, named Aries. Specifically, Aries has two sophisticated modules: (1) An novel trajectory embedding strategy enhanced by the bidirectional LSTM encoder and spatial attention mechanism, which can extract more precise and comprehensive knowledge. (2) A deep metric learning network aggregating multiple measures for better top-k query. Extensive experiments conducted on real trajectory dataset show that Aries achieves both impressive accuracy and lower training time compared with state-of-the-art solutions. In particular, it achieves 5x-10x speedup and 10%-20% accuracy improvement over Euclidean, Hausdorff, DTW, and EDR measures. Besides, our method can maintain stable performance when handling various scenarios, without repeated training in order to adapt to diverse similarity metrics.

In the edge computing Internet of Things network, we minimize the offloading overhead (caused by the bandwidth cost for data transmission and computation resource consumption for task remote processing) while providing the end-to-end (E2E) delay provisioning. Under the scenario, a tandem queue consisting of a transmission queue and a computing process queue is formed by the tasks offloaded to the edge server via wireless link and then processed through the computing resource. Due to the tandem queue, the offloading decision and computing resource allocation are coupled over the tandem queue. To make the problem tractable, we decouple the above two operations and propose a two-stage offloading filtering and computing resource allocation policy. After decouple, we then investigate the delay bound violation probability of the tandem queue by leveraging large deviation analysis. Further, we reveal that under the same E2E delay provisioning, the offloading overhead under the proposed decoupled policy can approach to the non-decoupled optimum by selecting an appropriate value of control parameter. Simulation results verify the theoretical analysis and show the efficiency of the proposed policy.

Abstract Energy harvesting (EH) aided Internet of Things (IoT) network is a promising paradigm to librate IoT network from energy deficiency. Dynamic energy and traffic scheduling in such a scenario is challenging due to temporal correlation of energy constraints and delay requirements of IoT applications. In this paper, joint energy management and sampling rate control to explore the tradeoff between network utility and delay performance are studied while maintaining the energy causality constraint. Taking into account the dynamic characteristics of EH process, channel fading and traffic arrivals, a stochastic optimisation problem is formulated to maximise the network utility. Leveraging the Lyapunov optimisation approach, combined with the idea of weight perturbation, a framework is proposed to decompose the stochastic problem into several deterministic sub‐problems that can be solved separately. Based on the framework, an online resource allocation algorithm is developed to achieve two major goals: first, balancing energy consumption and energy harvesting to stabilise their data and energy queues; second, deriving the utility‐delay tradeoff by adjusting the control parameter. The stability of data buffer and energy buffer in the proposed network is theoretical verified with performance analysis.

Non-orthogonal multiple access (NOMA) is considered as a potential technique to relieve the congestion due to concurrent access from massive devices in machine-to-machine (M2M) communication system. However, the cochannel interference caused by NOMA, and the energy budget of machine-type devices (MTDs), become the bottleneck to further improve the system performance. Given above issues, we formulate the joint user scheduling and energy management problem as a stochastic optimization problem. Specifically, the goal of the problem is to maximize the long-term average sum rate under the constraint of all MTDs’ quality-of-service (QoS) requirements. For tractability, the stochastic problem is firstly transformed into two static subproblems based on Lyapunov optimization. Then, using successive convex approximation (SCA) method, we design an effective algorithm to deal with the joint user scheduling and power allocation subproblems, which is a mixed integer and non-convex programming (MINCP). Simulation results demonstrate that our proposed algorithm has a good performance in convergence and outperforms other schemes in terms of user satisfaction.

The detection of frame deletion forgery is of great significance in the field of video forensics. Existing approaches, however, are not applicable to video sequences with variable motion strengths. In addition, the impact of interfering frames has not been considered in these approaches. Our research aims to develop a motion-adaptive forensic method as well as to eliminate interfering frames. Through a study of the statistical characteristics of the most common interfering frames such as relocated I-frames, we develop a new fluctuation feature based on frame motion residuals to identify frame deletion points (FDPs). The fluctuation feature is further enhanced by an intra-prediction elimination procedure so that it can be adapted to sequences with various motion levels. The enhanced feature is measured using a moving window detector to identify the location of a FDP. Finally, a postprocessing procedure is proposed to eliminate the minor interferences of sudden lighting change, focus vibration, and frame jitter. Our experimental results demonstrate that for videos with variable motion strengths and different interfering frames, the true positive rate of the algorithm can reach 90% when the false alarm rate is 0.3%. Our proposed method could provide a foundation for many practical applications of video forensics.

Detection of frame deletion is of great significance in the field of video forensics. Several approaches have been presented through analyzing the side effect caused by frame deletion. However, most of the current approaches can detect the existence of frame deletion but not the exact location of it. In this paper, we present a method which can directly locate the frame deletion point. Through the analysis of the distinguishing fluctuation feature of motion residual caused by frame deletion compared to interference frames and ordinary video content jitter in tampered video sequence, an algorithm based on the total motion residual of video frame is proposed to detect the frame deletion point. Moreover, an initiative processing procedure for frame motion residual and an adaptive threshold detector are introduced so that the robustness of the detection can be markedly improved. Experimental results show that the proposed algorithm is effective in generalized scenarios such as different encoding settings, rapid or slow motion sequences and multiple group of picture deletion. It also has a high performance that the true positive rate reaches 90% and the false alarm rate is less than 0.8%.

Remote sensing image enhancement occupies a peculiar position in remote sensing image processing and is an important preprocessing step for subsequent analysis. Numerous image enhancement techniques are available for remote sensing image enhancement. In this paper, the color constancy technique is introduced, and a novel color constancy remote sensing images enhancement algorithm is proposed. This algorithm can not only restore more details in the dark area of the image, but also self-adaptive to the luminance conditions. Based on the linear transform, the proposed algorithm contains two parts: (1) the scale parameter is calculated by the adaptive quadratic function with gamma correction to enhance the luminance; (2) the shifting parameter is used to restore the edge details. The experiments are conducted using images downloaded from NASA's website. Experimental results indicated that the proposed algorithm performs much better in preserving the hue and saturation and avoiding color distortion, especially in the dark area.