With the breakthroughs in deep learning, recent years have witnessed a booming of artificial intelligence (AI) applications and services, spanning from personal assistant to recommendation systems video/audio surveillance. More recently, with proliferation mobile computing Internet Things (IoT), billions IoT devices are connected Internet, generating zillions bytes data at network edge. Driving by this trend, there is an urgent need push AI frontiers edge so as fully unleash potential big...

As a key technology of enabling Artificial Intelligence (AI) applications in 5G era, Deep Neural Networks (DNNs) have quickly attracted widespread attention. However, it is challenging to run computation-intensive DNN-based tasks on mobile devices due the limited computation resources. What's worse, traditional cloud-assisted DNN inference heavily hindered by significant wide-area network latency, leading poor real-time performance as well low quality user experience. To address these...

Recent advances in artificial intelligence have driven increasing intelligent applications at the network edge, such as smart home, factory, and city. To deploy computationally intensive Deep Neural Networks (DNNs) on resource-constrained edge devices, traditional approaches relied either offloading workload to remote cloud or optimizing computation end device locally. However, cloud-assisted suffer from unreliable delay-significant wide-area network, local computing are limited by...

With the revolution of smart industry, more and Industrial Internet Things (IIoT) devices as well AI algorithms are deployed to achieve industrial intelligence. While applying computation-intensive deep learning on IIoT devices, however, it is challenging meet critical latency requirement for manufacturing. Traditional wisdom resorts cloud-centric paradigm but still works either inefficiently or ineffectively due heavy transmission overhead. To address this challenge, we propose Boomerang,...

Mobile-edge computing (MEC) has emerged as a promising supporting architecture providing variety of resources to the network edge, thus acting an enabler for edge intelligence services empowering massive mobile and Internet-of-Things (IoT) devices with artificial (AI) capability. With assistance servers, user equipments (UEs) are able run deep neural (DNN)-based AI applications, which generally resource hungry computation intensive such that individual UE can hardly afford by itself in real...

With the wide penetration of smart robots in multifarious fields, simultaneous localization and mapping (SLAM) technique robotics has attracted growing attention community. Yet collaborating SLAM over multiple still remains challenging due to performance contradiction between intensive graphics computation limited computing capability robots. While traditional solutions resort powerful cloud servers acting as an external provider, we show by real-world measurements that significant...

Recent years have witnessed a thriving growth of computing facilities connected at the network edge, cultivating edge networks as fundamental infrastructure for supporting miscellaneous intelligent services. Meanwhile, Artificial Intelligence (AI) frontiers extrapolated to graph domain and promoted Graph (GI). Given inherent relation between graphs networks, interdiscipline learning i.e., Edge GI or EGI, has revealed novel interplay them – aids in optimizing while facilitate model...

Big artificial intelligence (AI) models have emerged as a crucial element in various intelligent applications at the edge, such voice assistants smart homes and autonomous robotics factories. Training big AI – for example, personalized fine-tuning continual model refinement poses significant challenges to edge devices due inherent conflict between limited computing resources intensive workload associated with training. Despite constraints of on-device training, traditional approaches usually...

Edge intelligence has arisen as a promising computing paradigm for supporting miscellaneous smart applications that rely on machine learning techniques. While the community extensively investigated multi-tier edge deployment traditional deep models (e.g. CNNs, RNNs), emerging Graph Neural Networks (GNNs) are still under exploration, presenting stark disparity to its broad adoptions such traffic flow forecasting and location-based social recommendation. To bridge this gap, paper formally...

On-device Deep Neural Network (DNN) training has been recognized as crucial for privacy-preserving machine learning at the edge. However, intensive workload and limited onboard computing resources pose significant challenges to availability efficiency of model training. While existing works address these through native resource management optimization, we instead leverage our observation that edge environments usually comprise a rich set accompanying trusted devices with idle beyond single...

With the breakthroughs in deep learning, recent years have witnessed a booming of artificial intelligence (AI) applications and services, spanning from personal assistant to recommendation systems video/audio surveillance. More recently, with proliferation mobile computing Internet-of-Things (IoT), billions IoT devices are connected Internet, generating zillions Bytes data at network edge. Driving by this trend, there is an urgent need push AI frontiers edge so as fully unleash potential big...

To meet the stringent requirement of artificial intelligence applications, such as face recognition and video streaming analytics, a resource-constrained device can offload its task to nearby resource-rich devices in edge computing. Resource awareness, prime prerequisite for offloading decision-making, is critical achieving efficient collaborative computation performance. In this paper, we consider cost-aware resource probing (CERP) framework design infrastructure-free computing wherein...

Graph Neural Networks (GNNs) have gained growing interest in miscellaneous applications owing to their outstanding ability extracting latent representation on graph structures. To render GNN-based service for IoT-driven smart applications, the traditional model serving paradigm resorts cloud by fully uploading geo-distributed input data remote datacenter. However, our empirical measurements reveal significant communication overhead of such cloud-based and highlight profound potential...

Federated Learning (FL) has been a promising paradigm in distributed machine learning that enables in-situ model training and global aggregation. While it can well preserve private data for end users, to apply efficiently on IoT devices yet suffer from their inherent variants: available computing resources are typically constrained, heterogeneous, changing dynamically. Existing works deploy FL by pruning sparse or adopting tiny counterpart, which alleviates the workload but may have negative...

Federated Learning (FL) has emerged as a fundamental learning paradigm to harness massive data scattered at geo-distributed edge devices in privacy-preserving way. Given the heterogeneous deployment of devices, however, their are usually Non-IID, introducing significant challenges FL including degraded training accuracy, intensive communication costs, and high computing complexity. Towards that, traditional approaches typically utilize adaptive mechanisms, which may suffer from scalability...

Large language models (LLMs) have unlocked a plethora of powerful applications at the network edge, such as intelligent personal assistants. Data privacy and security concerns prompted shift towards edge-based fine-tuning LLMs, away from cloud reliance. However, this raises issues computational intensity resource scarcity, hindering training efficiency feasibility. While current studies investigate parameter-efficient (PEFT) techniques to mitigate constraints, our analysis indicates that...

With wide penetration of smart robots in many application fields, Simultaneous Localization And Mapping (SLAM) has attracted great attention the community. Yet performance issue on multi-robot SLAM still remains challenging due to contradiction constrained on-device resources and intensive graphics computation. While traditional approaches resort powerful cloud servers accelerate computing, we show by real-world measurements that significant communication overhead prevents its practicability...

Graph Neural Networks (GNNs) have gained growing interest in miscellaneous applications owing to their outstanding ability extracting latent representation on graph structures. To render GNN-based service for IoT-driven smart applications, traditional model serving paradigms usually resort the cloud by fully uploading geo-distributed input data remote datacenters. However, our empirical measurements reveal significant communication overhead of such cloud-based and highlight profound...

Accurate navigation is of paramount importance to ensure flight safety and efficiency for autonomous drones. Recent research starts use Deep Neural Networks (DNN) enhance drone given their remarkable predictive capability visual perception. However, existing solutions either run DNN inference tasks on drones in-situ, impeded by the limited onboard resource, or offload computation external servers which may incur large network latency. Few works consider jointly optimizing offloading...

To identify dense and small-size pedestrians in surveillance systems, high-resolution cameras are widely deployed, where images captured delivered to off-the-shelf pedestrian detection models. However, given the highly computation-intensive workload brought by high resolution, resource-constrained fail afford accurate inference real time. address that, we propose Hode, an offloaded video analytic framework that utilizes multiple edge nodes proximity expedite with inputs. Specifically, Hode...

Accurate navigation is of paramount importance to ensure flight safety and efficiency for autonomous drones. Recent research starts use Deep Neural Networks (DNN) enhance drone given their remarkable predictive capability visual perception. However, existing solutions either run DNN inference tasks on drones in situ, impeded by the limited onboard resource, or offload computation external servers which may incur large network latency. Few works consider jointly optimizing offloading...