The rapid advancement of artificial intelligence (AI) technologies has led to an increasing deployment of AI models on edge and terminal devices, driven by the proliferation of the Internet of Things (IoT) and the need for real-time data processing. This survey comprehensively explores the current state, technical challenges, and future trends of on-device AI models. We define on-device AI models as those designed to perform local data processing and inference, emphasizing their characteristics such as real-time performance, resource constraints, and enhanced data privacy. The survey is structured around key themes, including the fundamental concepts of AI models, application scenarios across various domains, and technical challenges faced in edge environments. We also discuss optimization and implementation strategies, such as data preprocessing, model compression, and hardware acceleration, which are essential for effective deployment. Furthermore, we examine the impact of emerging technologies, including edge computing and foundation models, on the evolution of on-device AI models. By providing a structured overview of the challenges, solutions, and future directions, this survey aims to facilitate further research and application of on-device AI, ultimately contributing to the advancement of intelligent systems in everyday life.

Fog Computing (FC) is a flexible architecture to support distributed domain-specific applications with cloud-like quality of service. However, current FC still lacks the mobility support mechanism when facing many mobile users with diversified application quality requirements. Such mobility support mechanism can be critical such as in the industrial internet where human, products, and devices are moveable. To fill in such gaps, in this paper we propose novel container migration algorithms and architecture to support mobility tasks with various application requirements. Our algorithms are realized from three aspects: 1) We consider mobile application tasks can be hosted in a container of a corresponding fog node that can be migrated, taking the communication delay and computational power consumption into consideration; 2) We further model such container migration strategy as multiple dimensional Markov Decision Process (MDP) spaces. To effectively reduce the large MDP spaces, efficient deep reinforcement learning algorithms are devised to achieve fast decision-making and 3) We implement the model and algorithms as a container migration prototype system and test its feasibility and performance. Extensive experiments show that our strategy outperforms the existing baseline approaches 2.9%, 48.5% and 58.4% on average in terms of delay, power consumption, and migration cost, respectively.

Due to the features of lightweight and easy deployment, the use of containers has emerged as a promising approach for Mobile Edge Computing (MEC). Before running the container, an image composed of several layers must exist locally. However, it has been conspicuously neglected by existing work that task scheduling at the granularity of the layer instead of the image can significantly reduce the task completion time to further meet the real-time requirement and resource efficiency in resource-limited MEC. To bridge the gap, considering the complex dependency between layers and images, a novel layer dependency-aware container scheduling algorithm is proposed to reduce the total task completion time. Specifically: 1) We model the online layer dependency-aware scheduling problem for containers in a heterogeneous MEC, considering the layer download time and task computation time. 2) A policy gradient algorithm is proposed to solve this problem, and the high-dimensional and low-dimensional relations for layer dependencies are extracted with improved action selection. 3) Experiments based on the real-world data trace show that the proposed algorithm outperforms the image-based and layer-based baseline algorithms by 54% and 19% on average, respectively.

Containers are becoming a popular way of running applications in edge computing. Before running the application, the edge node must download the application's container image consisting of multiple layers. However, given the limited bandwidth in edge computing, the container startup latency due to long image download time seriously affects the real-time performance. In this paper, we jointly determine the container assignment and the layer download sequence to reduce the total startup latency. We formulate the Container Assignment and Layer Sequencing (CALS) problem and prove its NP-hardness. A Layer-Aware Scheduling Algorithm (LASA) is proposed, fully considering layer sharing among images. Firstly, layers shared by the same set of images are grouped to reduce CALS's problem scale without affecting the optimal result. Secondly, considering both layer sharing and existing layer size on edge nodes, a layer-aware algorithm is designed to assign containers to appropriate edge nodes. Finally, to determine the layer download sequence on each edge node, an approximation algorithm is proposed. We further analyze the approximation ratio of LASA in the case of identical edge nodes with sufficient capacity. Extensive experiments based on real-world data show the effectiveness of LASA, which reduces the total startup latency by 40\% to 60\%.

Recently, many deep reinforcement learning (DRL)-based task scheduling algorithms have been widely used in edge computing (EC) to reduce energy consumption. Unlike the existing algorithms considering fixed and fewer edge nodes (servers) and tasks, in this article, a representation model with a DRL based algorithm is proposed to adapt the dynamic change of nodes and tasks and solve the dimensional disaster in DRL caused by a massive scale. Specifically, 1) we apply the representation learning models to describe the different nodes and tasks in EC, i.e., nodes and tasks are mapped to corresponding vector sub-spaces to reduce the dimensions and store the vector space efficiently. 2) With the space after dimensionality reduction, a DRL-based algorithm is employed to learn the vector representations of nodes and tasks and make scheduling decisions. 3) The experiments are conducted with the real-world data set, and the results show that the proposed representation model with DRL-based algorithm outperforms the baselines 18.04 and 9.94 percent on average regarding energy consumption and service level agreement violation (SLAV), respectively.

Driver fatigue is a leading factor in road accidents that can cause severe fatalities. Existing fatigue detection works focus on vision and electroencephalography(EEG) based means of detection. However, vision-based approaches suffer from view-blocking or vision distortion problems and EEG-based systems are intrusive, and the drivers have to use/wear the devices with inconvenience or additional costs. In our work, we propose a novel Wi-Fi signals based fatigue detection approach, called WiFind to overcome the drawbacks as associated with the current works. WiFind is simple and (wearable) device-free. It can detect the fatigue symptoms in the vehicle without relying on any visual image or video. By applying self-adaptive method, it can recognize the body features of drivers in multiple modes. It applies Hilbert-Huang transform(HHT) based pattern extract method results in accuracy increase in motion detection mode. WiFind can be easily deployed in a commodity Wi-Fi infrastructure, and we have evaluated its performance in real driving environments. The experimental results have shown that WiFind can achieve the recognition accuracy of 89.6% in a single driver scenario.