Convergence of Edge Computing and Deep Learning: A Comprehensive Survey (1907.08349v3)

Abstract: Ubiquitous sensors and smart devices from factories and communities are generating massive amounts of data, and ever-increasing computing power is driving the core of computation and services from the cloud to the edge of the network. As an important enabler broadly changing people's lives, from face recognition to ambitious smart factories and cities, developments of artificial intelligence (especially deep learning, DL) based applications and services are thriving. However, due to efficiency and latency issues, the current cloud computing service architecture hinders the vision of "providing artificial intelligence for every person and every organization at everywhere". Thus, unleashing DL services using resources at the network edge near the data sources has emerged as a desirable solution. Therefore, edge intelligence, aiming to facilitate the deployment of DL services by edge computing, has received significant attention. In addition, DL, as the representative technique of artificial intelligence, can be integrated into edge computing frameworks to build intelligent edge for dynamic, adaptive edge maintenance and management. With regard to mutually beneficial edge intelligence and intelligent edge, this paper introduces and discusses: 1) the application scenarios of both; 2) the practical implementation methods and enabling technologies, namely DL training and inference in the customized edge computing framework; 3) challenges and future trends of more pervasive and fine-grained intelligence. We believe that by consolidating information scattered across the communication, networking, and DL areas, this survey can help readers to understand the connections between enabling technologies while promoting further discussions on the fusion of edge intelligence and intelligent edge, i.e., Edge DL.

Convergence of Edge Computing and Deep Learning: A Comprehensive Survey

Edge computing and deep learning (DL) are two rapidly evolving fields that are intersecting to meet the demands of modern data-driven applications. The paper "Convergence of Edge Computing and Deep Learning: A Comprehensive Survey" provides an extensive review of this intersection. This document delineates how the proximity of edge computing resources to data sources can alleviate the inherent challenges in deep learning applications, such as bandwidth consumption and latency. The survey introduces the concept of "edge intelligence," where AI computations, particularly DL, are pushed to the edge of the network.

Motivations and Challenges

The motivation behind combining edge computing with deep learning stems from the increasing amount of data generated by IoT devices and the need for rapid processing of this data. Traditional cloud-based models suffer from significant latency and bandwidth issues, especially as the number of devices increases. Moreover, privacy concerns arise when sensitive data must be continuously transmitted to centralized servers. The convergence of edge computing and DL offers a solution by processing data locally at the edge, thereby reducing latency, saving bandwidth, and protecting privacy.

Application Scenarios

The paper examines several application scenarios where integrating edge computing with DL can be particularly beneficial:

1. Real-time Video Analytics: By deploying DL models locally on edge devices or edge nodes, the latency can be significantly reduced compared to cloud-based solutions. This is crucial for applications like smart surveillance and autonomous vehicles, where decisions must be made in real-time.
2. Autonomous Internet of Vehicles (IoVs): Leveraging edge computing can optimize the complex systems required for connected vehicle applications, improving safety and efficiency through rapid, localized data processing.
3. Intelligent Manufacturing: Edge computing facilitates real-time monitoring and predictive maintenance in smart factories, reducing downtime and enhancing operational efficiency.
4. Smart Homes and Cities: By embedding intelligence at the edge, applications such as smart lighting and energy management systems can operate more efficiently and with lower latency, offering better services and user experiences.

Deep Learning Models and Techniques

The paper discusses various DL models pertinent to edge computing, including CNNs, RNNs, GANs, and reinforcement learning techniques like Deep Q-Learning (DQL) and Policy Gradient methods. It emphasizes the need for optimized models considering the edge's resource constraints. Techniques such as model pruning, quantization, and knowledge distillation are explored to reduce the size and complexity of DL models to make them suitable for edge environments.

Edge Computing Architectures for DL

The survey outlines different edge computing paradigms, including Cloudlets, Mobile Edge Computing (MEC), and Fog Computing, detailing their roles in supporting DL. These architectures are designed to bring computing capabilities closer to the data source, thus reducing the latency and improving the efficiency of DL computations.

Hardware for Edge Computing

To support DL at the edge, specialized hardware platforms like NVIDIA's Jetson, Intel's Movidius, and FPGAs are discussed. These platforms are optimized for DL inference and training tasks under the resource constraints typical of edge environments.

Frameworks

Frameworks such as KubeEdge and OpenEdge are examined for orchestrating edge computing resources and deploying DL models efficiently. These frameworks facilitate the management and execution of lightweight containerized applications, which is essential for the dynamic and resource-constrained nature of edge environments.

Challenges and Future Directions

The paper identifies several challenges in deploying DL at the edge:

1. Model Optimization: Continuous research is needed to develop optimization techniques that can effectively balance model accuracy and resource consumption.
2. Collaborative Inference: Exploring more sophisticated methods for partitioning DL models between the end devices, edge, and cloud to improve performance and efficiency.
3. Standardization and Interoperability: Developing standardized protocols and APIs to ensure compatibility and seamless integration across different edge devices and platforms.
4. Scalability: Ensuring that the edge computing infrastructure can scale to cope with the growing number of devices and the increasing volume of data.
5. Security and Privacy: Implementing robust security measures to protect sensitive data and ensure privacy, especially in distributed and heterogeneous environments typical of edge computing.

Implications and Conclusion

The integration of DL with edge computing presents profound implications for various fields, enabling more efficient, real-time, and privacy-preserving AI applications. As edge devices continue to grow in computational capability, the edge will become an increasingly vital component of the AI pipeline. The survey sets a foundation for future research and development in this area, encouraging further exploration into optimizing edge computing frameworks and DL models for diverse and dynamic application scenarios.