Abstract:Deep learning technologies have demonstrated remarkable effectiveness in a wide range of tasks, and deep learning holds the potential to advance a multitude of applications, including in edge computing, where deep models are deployed on edge devices to enable instant data processing and response. A key challenge is that while the application of deep models often incurs substantial memory and computational costs, edge devices typically offer only very limited storage and computational capabilities that may vary substantially across devices. These characteristics make it difficult to build deep learning solutions that unleash the potential of edge devices while complying with their constraints. A promising approach to addressing this challenge is to automate the design of effective deep learning models that are lightweight, require only a little storage, and incur only low computational overheads. This survey offers comprehensive coverage of studies of design automation techniques for deep learning models targeting edge computing. It offers an overview and comparison of key metrics that are used commonly to quantify the proficiency of models in terms of effectiveness, lightness, and computational costs. The survey then proceeds to cover three categories of the state-of-the-art of deep model design automation techniques: automated neural architecture search, automated model compression, and joint automated design and compression. Finally, the survey covers open issues and directions for future research.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the application of design automation techniques in constructing fast, lightweight, and effective deep - learning models, especially the challenges in edge - computing environments. Specifically: 1. **Resource Constraints**: Edge devices usually have very limited storage and computing capabilities, which contradicts the fact that deep - learning models often require a large amount of memory and computing resources. How to maintain the effectiveness and performance of the model while satisfying the hardware limitations of edge devices is a key challenge. 2. **Design Complexity**: Building efficient deep - learning models not only requires advanced professional knowledge and experience, but also, due to the diversity of different mobile platforms and tasks, manually designing specific models to meet these requirements is both time - consuming and inconvenient. Moreover, the manual - design method has limitations in fully exploiting the hardware potential. To address the above challenges, the paper explores optimizing the design of deep - learning models through automated - design techniques, including Automated Neural Architecture Search, Automated Model Compression, and Joint Automated Design and Compression methods. These techniques aim to reduce manual labor, improve model efficiency, and ensure the accuracy and performance of the model at the same time. ### Main Contributions of the Paper 1. **Comprehensive Review**: The paper provides a comprehensive review of design - automation techniques for fast, lightweight, and effective deep - learning models, covering the analysis and comparison of more than 150 related literatures. 2. **New Taxonomy**: A new taxonomy is proposed to classify existing design - automation methods from multiple perspectives such as design methods (search, compression, or joint search and compression), design objectives (search space, search strategy, performance - evaluation strategy), and compression objects (tensors, knowledge, representation). 3. **Evaluation Metrics**: Various evaluation metrics for evaluating models and design methods are summarized and compared, emphasizing the role of each metric as well as its advantages and disadvantages. 4. **Future Directions**: Open problems in current research are discussed, and future research directions are pointed out, aiming to accelerate the further development of this field. Through these contributions, the paper provides a systematic reference framework for researchers and engineers, helping them better understand and apply design - automation techniques, thereby promoting the development of efficient deep - learning models.

Design Automation for Fast, Lightweight, and Effective Deep Learning Models: A Survey

Deep Learning on Mobile and Embedded Devices: State-of-the-art, Challenges, and Future Directions

Deep Learning in the Era of Edge Computing: Challenges and Opportunities

Survey on Energy-Efficient Deep Neural Networks for Computer Vision

Design Automation for Efficient Deep Learning Computing

Application of Deep Learning in Back-End Simulation: Challenges and Opportunities

Computation-efficient Deep Learning for Computer Vision: A Survey

Lightweight Deep Learning for Resource-Constrained Environments: A Survey

Deep Model Compression for Mobile Platforms: A Survey

Model Compression for Deep Neural Networks: A Survey

Understanding Sensor Data Using Deep Learning Methods on Resource-Constrained Edge Devices.

A Survey on Hardware Accelerator Design of Deep Learning for Edge Devices

Efficient Deep Learning Infrastructures for Embedded Computing Systems: A Comprehensive Survey and Future Envision

Enable Deep Learning on Mobile Devices: Methods, Systems, and Applications

Deep Learning for Edge Computing Applications: A State-of-the-Art Survey

Resource-Efficient Deep Learning: A Survey on Model-, Arithmetic-, and Implementation-Level Techniques

Large-Scale Deep Learning Optimizations: A Comprehensive Survey

From Algorithm to Hardware: A Survey on Efficient and Safe Deployment of Deep Neural Networks

On-Demand Deep Model Compression for Mobile Devices

Bringing AI To Edge: From Deep Learning's Perspective