Digital Twin (DT) technology is a key enabler of the Industrial Internet of Things (IIoT), facilitating predictive control, fault detection, and simulation through high-fidelity virtual replicas of physical assets. To alleviate the significant computational burden on the Cloud Server (CS) of centralized DT services, existing solutions commonly deploy DT modules on edge servers (ESs). However, deploying each DT module at the edge requires intensive resources to support frequent data updates, processing, and analysis; thus, unrestricted module deployment can pose substantial sustainability challenges, particularly when renewable energy availability is limited. To tackle this challenge, we propose a green DT-IIoT architecture and formulate a DT module placement problem that jointly minimizes DT service freshness and carbon emissions, subject to constraints on data synchronization accuracy, computing resources, and storage capacity. Among these, DT service freshness, which indicates real-time responsiveness, and data synchronization accuracy, which reflects the reliability of real-time input data, are two critical metrics affecting DT service quality. Furthermore, we propose a Dueling Double Deep Q-Network (D3QN) based placement algorithm (DDMP), which achieves high performance with a relatively simple structure that is well-suited for rapidly evolving IIoT scenarios. Simulation results demonstrate that our proposed approach enhances DT service freshness while reducing carbon emissions compared with baseline methods.

The exponential increase in connected user devices poses scalability challenges for centralized cloud computing. Mobile Edge Computing (MEC) and Fog Computing alleviate latency by deploying computation and storage resources closer to end-users. However, due to the resource limitations, heterogeneity, and dispersed nature of edge servers, there is a need to jointly optimize service caching and application placement strategies to enhance service quality. Given the widespread use of containerized services at the edge, we propose a distributed caching scheme that allows all edge nodes to cache services at the granularity of container image layers. This collaborative caching approach reduces the real-time latency, bandwidth consumption, and caching costs associated with retrieving and initializing applications. Additionally, to address the variability in application popularity across different edge regions, we model application popularity using a Zipf distribution and construct a multi-slot joint optimization model for caching and deployment decisions based on deployment cost, application startup time, and average delay. We then propose a two-stage optimization method to solve this model, demonstrating through comparison with centralized and P2P models the effectiveness of the proposed approach.

With the continuous advancement of network technologies and hardware devices, computation-intensive and latency-sensitive tasks have emerged worldwide, requiring networks to provide extensive coverage, low latency, and robust computing capabilities. Leveraging the global coverage of LowEarth Orbit (LEO) satellites and the flexible resource invocation capabilities of the Computing Power Network (CPN), we propose a Satellite-Terrestrial Computing Power Network (ST-CPN) framework that integrates both strengths. In this framework, tasks can be offloaded to satellites closer to users for processing, ensuring high-quality services anytime and anywhere. However, due to the dynamic nature of the network and the limited resources of individual nodes, efficiently executing complex dependent tasks presents significant challenges. Therefore, we investigate the dependent task offloading problem in the dynamic ST-CPN environment. Considering dynamic changes of topology and available resources caused by satellite mobility, we propose a Time-Space-Varying Resource Graph (TSVRG) to capture the status of the communication, storage, and computation resources. On this basis, given that individual nodes struggle to process dependent tasks, we offload multiple subtasks of a task to different nodes for collaborative processing. In this paper, we model the task as a Directed Acyclic Graph (DAG) and transform the offloading problem into a mapping problem from the DAG to TSVRG. We then introduce a Delay Predictionbased Graph Mapping Algorithm (DPGMA) to address this problem. Simulation results indicate that our scheme achieves better performance than the benchmark schemes.

Generative Artificial Intelligenge (GAI) is revolutionizing the world with its unprecedented content creation ability. Large Language Model (LLM) is one of its most embraced branches. However, due to LLM’s substantial size and resource-intensive nature, it is cloud-hosted, raising concerns about privacy, usage limitations, and latency. In this paper, we propose to utilize ubiquitous distributed wireless edge computing resources for real-time LLM inference. Specifically, we introduce a novel LLM edge inference framework, incorporating batching and model quantization to ensure high throughput inference on resource-limited edge devices. Then, based on the architecture of transformer decoder-based LLMs, we formulate an edge inference optimization problem which is NP-hard, considering batch scheduling and joint allocation of communication and computation resources. The solution is the optimal throughput under edge resource constraints and heterogeneous user requirements on latency and accuracy. To solve this NP-hard problem, we develop an OT-GAH (Optimal Tree-search with Generalized Assignment Heuristics) algorithm with reasonable complexity and$\frac {1}{2}$-approximation ratio. We first design the OT algorithm with online tree-pruning for single-edge-node multi-user case, which navigates the inference request selection within the tree structure to miximize throughput. We then consider the multi-edge-node case and propose the GAH algorithm, which recrusively invokes the OT in each node’s inference scheduling iteration. Simulation results demonstrate the superiority of OT-GAH batching over other benchmarks, revealing an over 45% time complexity reduction compared to brute-force searching.

Recently, the advancement of information technologies have accelerated the generation of big data, necessitating substantial computing power. This has spurred the development of Computing Power Networks (CPNs), which can overcome the limitations of computing power isolation. However, CPNs consume significant energy and produce large carbon emissions during big data processing. Therefore, an energy-efficient task scheduling scheme, coupled with the utilization of renewable energy, appears to be particularly necessary. Nevertheless, the interplay between computing power and networks, and the spatiotemporal variations in green CPNs pose a challenge to designing the task scheduling scheme. In this paper, we propose a transferable spatiotemporal task scheduling scheme with a triple selection of CPN nodes, routing paths, and forwarding time of tasks. The scheme can overcome the dynamics of green CPNs, and jointly optimize the energy consumption and carbon emissions with ensuring delay constraints and long-term load balancing. Then, we present a task scheduling algorithm based on improved nondominated sorting genetic algorithm-II (NSGA-II) to solve the problem, and numerical results demonstrate that our scheme is effective in reducing the overall energy consumption and carbon emissions of CPNs.

In recent years, there has been a proliferation of Edge Intelligence (EI) services, especially within Internet of Vehicles (IoV) scenarios, accompanied by a growing demand for multi-modal content generation. In response, Generative Artificial Intelligence (GAI) has emerged as a promising solution, equipping EI to produce diverse Artificial Intelligence-Generated Content (AIGC) for ubiquitous edge services. However, existing cloud-based GAI capabilities, which are mostly provided via the web and the Internet, introduce unacceptable latency overhead and heightened security risks for vehicular services. To address the above shortcomings and the lack of endogenous mechanisms for applying GAI to IoV scenarios, in this paper, we propose a layered vehicular GAI framework that seamlessly integrates GAI and EI. Within this framework, we devise a distributed collaborative inference mechanism between Road-Side Units (RSUs) and vehicles. Furthermore, we formulate the shared and local inference splitting problem, a pivotal challenge influencing both GAI service latency and content-generation capability. To tackle this issue, we introduce a backward induction-based algorithm, which enables the system can make splitting decisions using a simple threshold-based policy. Simulation results underscore the remarkable performance of the proposed system and vehicular collaborative inference mechanism, promising to facilitate diverse content generation within vehicular networks.

The utilization of emergent Generative Artificial Intelligence (GAl) within the realm of Internet of Vehicles (loV) can augment edge intelligence, thereby catering to the diverse content-generation needs of novel in-vehicle services. Nonethe-less, existing cloud-centric GAl paradigms are not inherently suitable for wireless vehicular networks, primarily due to their extensive computing requirements, lack of specificity, and spatial detachment from end users. To cope with these challenges, we introduce an innovative Generative 10 V (g 10 V)architecture that employs a collaborative fine-tuning mechanism for pre-trained GAl models. The mechanism is mainly orchestrated collaboratively by Road-Side Units (RSUs) and vehicles within a Federated Learning (FL) paradigm. Here, we take text-to-image diffusion models as typical examples to show the co-fine-tuning workflow in detail, aiming to utilize edge traffic data to realize rapid, customized, and lightweight GAl in the resource-limited 10 V scenario. Thereafter, we formulate the problem of edge communication and computation resource allocation during RSU-vehicle co-fine-tuning, which is pivotal for optimizing time and energy consumption within this process. To address the challenge, we deploy a Self-adaptive Harmony Search (SHS)-based resource allocation strategy. Experiments based on Stable Diffusion vl-4 model validate the excellent performance in image generating and the time and energy consumption during co-fine-tuning in resource-limited and fast-changing 10 V scenarios.

Generative Artificial Intelligence (GAI) is taking the world by storm with its unparalleled content creation ability. Large Language Models (LLMs) are at the forefront of this movement. However, the significant resource demands of LLMs often require cloud hosting, which raises issues regarding privacy, latency, and usage limitations. Although edge intelligence has long been utilized to solve these challenges by enabling real-time AI computation on ubiquitous edge resources close to data sources, most research has focused on traditional AI models and has left a gap in addressing the unique characteristics of LLM inference, such as considerable model size, auto-regressive processes, and self-attention mechanisms. In this paper, we present an edge intelligence optimization problem tailored for LLM inference. Specifically, with the deployment of the batching technique and model quantization on resource-limited edge devices, we formulate an inference model for transformer decoder-based LLMs. Furthermore, our approach aims to maximize the inference throughput via batch scheduling and joint allocation of communication and computation resources, while also considering edge resource constraints and varying user requirements of latency and accuracy. To address this NP-hard problem, we develop an optimal Depth-First Tree-Searching algorithm with online tree-Pruning (DFTSP) that operates within a feasible time complexity. Simulation results indicate that DFTSP surpasses other batching benchmarks in throughput across diverse user settings and quantization techniques, and it reduces time complexity by over 45% compared to the brute-force searching method.

The growth of intelligent vehicular services, like augmented reality (AR) road simulation, underscores the need for rapid, multi-modal content generation. Generative artificial intelligence (GAI) models, known for their swift production of diverse artificial intelligence-generated content (AIGC), stand out as a prime solution. However, integrating cloud-centric GAI models into vehicular networks is fraught with challenges. Notably, to offer specialized generative edge intelligence (EI) and boost vehicular AIGC, GAI models need to tap into user data and utilize significant computation resources. Moreover, their deployment across vehicular networks is essential for proximity-based distributed inferences. Yet, edge devices are resource-limited, and data sharing can raise safety and privacy concerns. Addressing these challenges, this paper introduces GAI-IoV, an EI-enabled GAI framework facilitated through the cooperation between road-side units (RSUs) and vehicles. Subsequently, we propose the workflow for collaborative fine-tuning and distributed inference. On this basis, two pivotal vehicle-centric problems are then formulated: computation and communication resource allocation for federated fine-tuning (FFT) to optimize time and energy cost, and splitting strategy of shared and local inferences to optimize inference latency and content-generation capability. To solve these optimizations, we introduce a self-adaptive global best harmony search (SGHS) algorithm for resource allocation and a backward induction method for determining inference splitting strategy. Our experiments based on the Stable Diffusion v1-4 model vouch for a superior fine-tuning and inference capabilities of GAI-IoV. Furthermore, simulations underscore its resource utilization and distributed inference efficiency in dynamic vehicular scenarios.

As Artificial Intelligence (AI) has flourished in various industries in recent years, the evolutionary trend of endogenous network intelligence continues to accelerate. Connected intelligence, which aims to achieve a widely distributed and collaborative evolution of intelligence, has received much attention. Meanwhile, the emerging Computing Power Network (CPN) provides more robust computation and communication capabilities for intelligence training and intelligent application processing. In this context, the integration of CPN and connected intelligence becomes a potential solution to drive the digital and intelligent transformation of networks. In this paper, we propose a scheme to jointly consider task scheduling, routing, and intelligence capability improvement during the processing of smart applications represented by Digital Twin (DT) in the CPN-enabled connected intelligence systems. We formulate the problem of jointly optimizing the processing time consumption and training accuracy improvement. We solve the problem using a modified NSGA-II algorithm and numerical results show that our approach is effective in optimizing the overall average time consumption and improving the accuracy of the intelligence models distributed in the system during task processing.