Graph neural networks (GNNs) have attracted significant attention in human action recognition (HAR) tasks due to their ability to model spatial-temporal relationships between body joints in skeletal graphs. However, prior solutions typically suffer from high computational overhead and significant inference latency, particularly on resource-constrained platforms like edge devices. To mitigate this, we propose Edge-STGNN, a speed-optimized neural architecture search (NAS) framework, for the automated design of spatial-temporal GNNs (STGNNs) tailored for action recognition on edge devices. Specifically, Edge-STGNN defines a search space that includes diverse temporal convolutional layers, attention mechanisms, and network depth options. Based on this search space, it constructs the supernet using a single-path training approach. Subsequently, an evolutionary algorithm-based search strategy is applied to identify optimal architectures. Furthermore, Edge-STGNN integrates an efficient speed predictor to reduce evaluation time, and employs a small-scale proxy dataset to lower search costs. Comprehensive experiments on multiple datasets and edge devices demonstrate that Edge-STGNN effectively identifies architectures with fewer parameters, reduced computational complexity, and faster inference speeds, while preserving acceptable recognition accuracy. Additionally, by incorporating speed constraints, Edge-STGNN enables a flexible trade-off between accuracy and speed tailored to specific application demands. For example, on the NTU RGB+D 60 xview60 benchmark, compared to all baseline methods with reported speeds, Edge-STGNN achieves a 2.4× to 9.4× inference speedup while maintaining accuracy within a ±1% margin.

Mobile edge computing (MEC) is a promising paradigm that provides abundant computation and storage resources at the edge close to mobile devices (MDs). In MEC networks, MDs offload compute-heavy tasks to nearby edge servers (ESs) for delay-sensitive processing, where relevant services are stored to support task execution. However, the limited computation and storage capacities of ESs make joint optimization of service caching and computation offloading challenging due to coupled decisions, a large solution space, and dynamic environments. In this paper, we investigate the joint optimization of service caching and computation offloading in MEC networks, aiming to maximize the cache hit ratio and minimize the average service latency. To tackle this problem, the original formulation is decomposed into two hierarchical subproblems, namely high-level service caching and low-level computation offloading. We propose a novel hierarchical deep reinforcement learning (DRL) algorithm with active inference, termed HADRL. At the high-level, we adopt a deep deterministic policy gradient (DDPG) based DRL approach to maximize the cache hit ratio. At the low-level, we employ an active inference based DRL approach to minimize the average service latency. Unlike conventional DRL, the active inference based DRL approach selects policies by minimizing expected free energy instead of relying only on explicit rewards, making it well suited for highly dynamic low-level computation offloading. According to the simulation outcomes, the HADRL scheme surpasses the benchmark algorithms with respect to cache hit ratio as well as average service latency.

This is a corrigendum for the article “LyDRL: Lyapunov-guided Deep Reinforcement Learning for Stable Task Offloading in Connected Autonomous Vehicles” published in ACM Trans. Autonom. Adapt. Syst. 20, 3, Article 24 (September 2025), 29 pages.

Task offloading is recognized as a promising approach to enhance the computational performance of Connected Autonomous Vehicles (CAVs). Some applications of CAVs, such as metaverse applications, require substantial resources, posing significant challenges to CAVs with limited computing and storage capacities. CAVs can offload resource-intensive applications to the Vehicular Edge Computing (VEC) server, which has strong computing capabilities. To fully utilize the resources in the CAV system, partial offloading is employed. However, the local computing resources are limited for continuously generated partial offloading tasks. This results in many partially locally executed tasks experiencing long processing times or being discarded, which is detrimental to delay-sensitive tasks on CAVs. This article proposes Lyapunov function-guided reinforcement learning for the CAVs task offloading computational framework, LyDRL. Specifically, LyDRL first uses the Lyapunov function to transform the long-term objective optimization problem into subproblems determined at each time slot. In each time slot, deep reinforcement learning is used to obtain the optimal task offloading decision while satisfying the constraints. Simulation results show that compared with the existing algorithms, the proposed strategy can ensure the stability of the CAVs system and achieve the lowest system overhead.

Deep neural networks (DNNs) have shown remarkable performance in the super-resolution (SR) task, which can upscale low-resolution images to satisfy application demands on image quality. However, the high computational intensity of DNN models poses a challenge to executing SR tasks on resource-constrained edge platforms. To leverage heterogeneous computational resources (e.g., CPU, GPU, and NPU) to speed up image reconstruction through concurrent inference, we propose a novel framework, called ESHP, for Efficient Super-resolution on edge platforms with Heterogeneous Processors. Our proposed ESHP framework boasts several advantageous characteristics: 1) it substantially speeds up SR processing over the existing approaches by leveraging all available heterogeneous hardware; 2) it uses deep reinforcement learning (DRL) to enable adaptive and optimal scheduling based on runtime states; 3) it strikes a balance between SR performance and computational cost during inference; and 4) it does not modify the original architecture of given SR model. We have conducted extensive experiments on typical edge platforms with popular SR models and resolution datasets of different scales, which verify the effectiveness and the versatility of our ESHP against other commonly-used baselines.

With the explosive growth in video streaming comes a rising demand for efficient and scalable video understanding. State-of-the-art video recognition approaches based on Convolutional Neural Network (CNN) have shown promising performance by adopting 2D or 3D CNN architectures. However, the large data volumes, high resource demands, and strict latency requirements have hindered the wide application of these solutions on resource-constrained Internet-of-Things (IoT) and edge devices. To address this issue, we propose a novel framework called TLEE that enables the input samples the abilities of both Temporal-wise and Layer-wise Early Exiting on 2D CNN backbones for efficient video recognition. TLEE consists of three types of modules: gating module, branch module, and feature reuse module. The gating module determines for an input video from which frame of this video to exit the per-frame computation, while the branch module determines for an input frame from which layer of the CNN backbone to exit the per-layer computation. Besides, based on the accumulated features of frame sequences from exit branches, the feature reuse module generates effective video representations to enable more efficient predictions. Extensive experiments on benchmark datasets demonstrate that the proposed TLEE can significantly outperform the state-of-the-art approaches in terms of computational cost and inference latency, while maintaining competitive recognition accuracy. In addition, we verify the superiority of TLEE on the typical edge device NVIDIA Jetson Nano.

Deep learning technology has grown significantly in new application scenarios such as smart cities and driverless vehicles, but its deployment needs to consume a lot of resources. It is usually difficult to execute inference task solely on resource-constrained Intelligent Internet-of-Things (IoT) devices to meet strictly service delay requirements. CNN-based inference task is usually offloaded to the edge server or cloud. However, it may lead to unstable performance and privacy leaks. To address the above challenges, this article aims to design a low latency distributed inference framework, EdgeCI, which assigns inference tasks to locally idle, connected, and resource-constrained IoT device cluster networks. EdgeCI exploits two key optimization knobs, including: (1) Auction-based Workload Assignment Scheme (AWAS), which achieves the workload balance by assigning each workload partition to the more matching IoT device; (2) Fused-Layer parallelization strategy based on non-recursive Dynamic Programming (DPFL), which is aimed at further minimizing the inference time. We have implemented EdgeCI based on PyTorch and evaluated its performance with VGG-16 and ResNet-34 image recognition models. The experimental results prove that our proposed AWAS and DPFL outperform the typical state-of-the-art solutions. When they are well combined, EdgeCI can improve inference speed by 34.72% to 43.52%. EdgeCI outperforms the state-of-the art approaches on our edge cluster.

Deep neural networks (DNNs) based approaches have achieved superior performance in single image super-resolution (SR). To obtain better visual quality, DNNs for SR are generally designed with massive computation overhead. To accelerate network inference under resource constraints, we propose a classification-based dynamic network for efficient super-resolution (CDNSR), which combines the classification and SR networks in a unified framework. Specifically, CDNSR decomposes a large image into a number of image-patches, and uses a classification network to categorize them into different classes based on the restoration difficulty. Each class of image-patches will be handled by the SR network that corresponds to the difficulty of this class. In particular, we design a new loss to trade off between the computational overhead and the reconstruction quality. Besides, we apply contrastive learning based knowledge distillation to guarantee the performance of SR networks and the quality of reconstructed images. Extensive experiments show that CDNSR significantly outperforms the other SR networks and backbones on image quality and computational overhead.

Model compression is a key technique that enables deploying Deep Neural Networks (DNNs) on Internet-of-Things (IoT) devices with constrained computing resources and limited power budgets. Channel pruning has become one of the representative compression approaches, but how to determine the compression ratio for different layers of a model still remains as a challenging task. Current automated pruning solutions address this issue by searching for an optimal strategy according to the target compression ratio. Nevertheless, when given a series of tasks with multiple compression ratios and different training datasets, these approaches have to carry out the pruning process repeatedly, which is inefficient and time-consuming. In this paper, we propose a Multi-Task Automated Channel Pruning (MTACP) framework, which can simultaneously generate a number of feasible compressed models satisfying different task demands for a target DNN model. To learn MTACP, the layer-by-layer multi-task channel pruning process is transformed into a Markov Decision Process (MDP), which seeks to solve a series of decision-making problems. Based on this MDP, we propose an actor-critic-based multi-task Reinforcement Learning (RL) algorithm to learn the optimal policy, working based on the IMPortance weighted Actor-Learner Architectures (IMPALA). IMPALA is known as a distributed RL architecture, in which the learner can learn from a set of actors that continuously generate trajectories of experience in their own environments. Extensive experiments on CIFAR10/100 and FLOWER102 datasets for MTACP demonstrate its unique capability for multi-task settings, as well as its superior performance over state-of-the-art solutions.

The advent of Deep Neural Networks (DNNs) has empowered numerous computer-vision applications. Due to the high computational intensity of DNN models, as well as the resource constrained nature of Industrial Internet-of-Things (IIoT) devices, it is generally very challenging to deploy and execute DNNs efficiently in the industrial scenarios. Substantial research has focused on model compression or edge-cloud offloading, which trades off accuracy for efficiency or depends on high-quality infrastructure support, respectively. In this article, we present EdgeDI, a framework for executing DNN inference in a partitioned, distributed manner on a cluster of IIoT devices. To improve the inference performance, EdgeDI exploits two key optimization knobs, including: (1) Model compression based on deep architecture design, which transforms the target DNN model into a compact one that reduces the resource requirements for IIoT devices without sacrificing accuracy; (2) Distributed inference based on adaptive workload partitioning, which achieves high parallelism by adaptively balancing the workload distribution among IIoT devices under heterogeneous resource conditions. We have implemented EdgeDI based on PyTorch, and evaluated its performance with the NEU-CLS defect classification task and two typical DNN models (i.e., VGG and ResNet) on a cluster of heterogeneous Raspberry Pi devices. The results indicate that the proposed two optimization approaches significantly outperform the existing solutions in their specific domains. When they are well combined, EdgeDI can provide scalable DNN inference speedups that are very close to or even much higher than the theoretical speedup bounds, while still maintaining the desired accuracy.

Nowadays, sensor-cloud systems have received wide attention from both academia and industry. Sensor-cloud system not only improves performances of wireless sensor networks (WSNs), but also combines different functional WSNs together to provide comprehensive services. However, a variety of malicious attacks threaten the sensor-cloud security, such as integrity, authenticity, availability and so on. Traditional available security mechanisms (e.g., cryptography and authentication) are still vulnerable. Although there are schemes to provide security by trust evaluation, the evaluation considers whether or not a sensor is credible only by checking the communication behaviors. Furthermore, when mobile sensor sinks are employed to collect sensing data, there appears a type of attacks called replicated sink attacks that are often ignored in the previous work. These attacks may bring serious vulnerability to trustworthy data collection in sensor-cloud systems. In this paper, we propose a comprehensive trustworthy data collection (CTDC) approach for sensor-cloud systems. Three kinds of trust, i.e., direct trust, indirect trust, and functional trust are defined to evaluate the trustworthiness of both sensors and mobile sinks. Except for resisting malicious attacks, the performances of sensor-cloud, such as energy, transmission distance and network throughput are also considered. We also conduct extensive simulations to evaluate the efficiency of CTDC. The simulation results show that CTDC correctly identifies malicious nodes and offers an improved performance in the data collection.

Motivated by the prospects of 5G communications and industrial Internet of Things (IoT), recent years have seen the rise of a new computing paradigm, edge computing, which shifts data analytics to network edges that are at the proximity of big data sources. Although deep neural networks (DNNs) have been extensively used in many platforms and scenarios, they are usually both compute and memory intensive, thus, difficult to be deployed on resource-limited edge devices and in performance-demanding edge applications. Hence, there is an urgent need for techniques that enable DNN models to fit into edge devices, while ensuring acceptable execution costs and inference accuracy. This article proposes an on-demand DNN model inference system for industrial edge devices, called knowledge distillation and early exit on edge (EdgeKE). It focuses on the following two design knobs: first, DNN compression based on knowledge distillation, which trains the compact edge models under the supervision of large complex models for improving accuracy and speed; second, DNN acceleration based on early exit, which provides flexible choices for satisfying distinct latency or accuracy requirements from edge applications. By extensive evaluations on the CIFAR100 dataset and across three state-of-art edge devices, experimental results demonstrate that EdgeKE significantly outperforms the baseline models in terms of inference latency and memory footprint, while maintaining competitive classification accuracy. Furthermore, EdgeKE is verified to be efficiently adaptive to the application requirements on the inference performance. The accuracy loss is within 4.84% under various latency constraints, and the speedup ratio is up to 3.30× under various accuracy requirements.

The recent emergence of mobile cloud computing has enabled mobile users to offload computing tasks from mobile devices to nearby cloudlets, so as to reduce energy consumption and improve application performance. In this paper, we consider the problem of maximizing the profit of the cloudlets' managing platform that receives computing requests from mobile users and fulfils these requests by leveraging computing service of participating cloudlets. However, it is very challenging to maximize the operating profit for such a managing platform, due to unpredictable arrival of user requests, dynamic participation of mobile cloudlets, and complexity in computing resource allocations. Based on the Lyapunov optimization technique combined with the technique of weight perturbation, we introduce a new stochastic control algorithm that makes online decisions on computing request admission and dispatching, computing service purchasing, and computing resource allocation. Different from traditional techniques, this algorithm does not require any statistical knowledge of relevant system dynamics, and is efficient for implementation in practice. Theoretical analysis and simulation results have demonstrated both the profit optimality and the system stability achieved by the proposed control algorithm.

Currently, the demand on portable medical electronic systems are increasing, for they provide services and information to both patients and doctors that the traditional medical methods cannot achieve. However, the development of portable medical electronic systems has many limitations, i.e., one has to find a sweet spot among performance, power consumption, size, and cost. For instance, if one only increases the memory of a system, not only does the cost go up, the power consumption also goes up, meanwhile the standby time goes down. In some cases, the hardware size goes up as well. One way of satisfying these constraints while retaining the design and functionality is to compress executable code and data as much as possible. In this paper, a novel clustered multi-dictionary code compression method is proposed to effectively reduce the memory size by replacing the most common codes with shorter codeword. The codes are clustered according to their repeating times. Each cluster is compressed with a different dictionary to make the codeword length different. By pairing clusters and dictionaries with the highest entropy, the compression efficiency becomes the best. Theoretical analysis and experimental results show that this method can achieve significant compression effect. The code of MiBench benchmark compiled under ARM and MIPS instruction set architecture are compressed with this method and the code size decreases by 50%. Aside from high compression ratio, our method also provides relatively fast encoding and very fast decoding.

This comment points out several mathematical errors in the proof of Therorem 3, and gives the correct expression of B3.

In , the authors proposed a distributed robust routing protocol RRP for mobile sensor networks based on node cooperation, and compared the energy efficiency of cooperative and non-cooperative routing through analysis and simulation. However, the model for energy consumption of sensor nodes is not correct. Therefore, we present a new energy analysis in this note.

Congestion in wireless sensor networks (WSNs) not only causes severe information loss but also leads to excessive energy consumption. To address this problem, a novel scheme for congestion avoidance, detection and alleviation (CADA) in WSNs is proposed in this paper. By exploiting data characteristics, a small number of representative nodes are chosen from those in the event area as data sources, so that the source traffic can be suppressed proactively to avoid potential congestion. Once congestion occurs inevitably due to traffic mergence, it will be detected in a timely way by the hotspot node based on a combination of buffer occupancy and channel utilization. Congestion is then alleviated reactively by either dynamic traffic multiplexing or source rate regulation in accordance with the specific hotspot scenarios. Extensive simulation results under typical congestion scenarios are presented to illuminate the distinguished performance of the proposed scheme.

Instant wireless sensor network (IWSN) is a type of WSN deployed for a class of special applications which have the common requirement on instantly responding for collecting and transmitting sensory data, e.g., volcanic eruption monitoring or nuclear leakage detection. In this paper, having a cluster-based WSN, we present a compensation-based reliable data delivery protocol (CRDD) to collect and transmit sensory data timely, reliably, and energy-efficiently. The CRDD consists of three important parts: (1) reliability calculating mechanism, (2) information classifying mechanism, and (3) intelligent balancing mechanism. By using these three mechanisms, the CRDD can reduce redundant messages for enhancing the transmission performance and compensate the deficient messages for reliability. The simulation results show that CRDD can outperform both LEACH and ECDG and significantly improve sensory data collection speed and system dependability.

Data forwarding in the wireless networks typically employs a sender-oriented approach in which the next hop node is pre-selected based on neighbor or network information. This method incurs large overhead when accurate information is needed for making the optimal forwarding decision. In this paper, a receiver-oriented robust data delivery scheme (RRDD) is proposed for mobile sensor networks. In RRDD, the sender does not appoint a specific forwarder proactively, but allows its neighboring candidates to dynamically contend for the data forwarding task based on local state information. In this way, the best-suited node is elected at each hop to provide robust and efficient delivery service to data packets. Comprehensive simulations show that RRDD exhibits superior transmission performance over all of the compared schemes.

Current schemes for Wireless Sensor Network addressing are mainly based on static sensor node address which is determined in the process of deployment. However, with the increase of the network size, it is getting difficult to allocate the addresses manually for a large number of nodes. Moreover, the static address scheme is not able to deal with the complex situation in the real world due to dynamic death and replenishment of the sensor nodes. Although some dynamic address allocation schemes have been proposed, the frequent control messages exchange used by these schemes will lead to both excessive drain of limited power supply and increased collisions in wireless communication. In this paper, we propose a new efficient address allocation scheme for query based sensor networks. The proposed scheme allocates a temporary network-wide unique address only to the sensor node which reports data in response to an explicit query from the sink. We have evaluated our scheme performance through both analysis and extensive simulation experiments. Our evaluation results show that the proposed scheme exhibits better performance than existing schemes.