Related Experiment Video
Updated: Jun 23, 2025
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.0K
From Algorithm to Hardware: A Survey on Efficient and Safe Deployment of Deep Neural Networks
Summary
This survey explores efficient deep neural network (DNN) deployment, covering model compression, hardware accelerators, and security. It aims to guide researchers toward high-performance, cost-effective, and safe AI systems.
Area of Science:
- Artificial Intelligence
- Computer Engineering
Background:
- Deep neural networks (DNNs) face deployment challenges due to high memory, energy, and computational costs.
- Model compression techniques like quantization and pruning are crucial for DNN efficiency.
- Integrating DNNs with hardware accelerators and ensuring security are key research areas.
Purpose of the Study:
- To provide a comprehensive survey of recent research on efficient and secure DNN deployment.
- To cover model compression, hardware acceleration, and security integration for DNNs.
- To offer a holistic view from algorithms to hardware and security.
Main Methods:
- Survey of mainstream model compression techniques (quantization, pruning, knowledge distillation).
- Review of hardware accelerators optimized for compressed DNNs.
- Discussion of security measures like homomorphic encryption for DNNs.
Main Results:
- Identified key model compression methods and their impact on DNN efficiency.
- Highlighted advancements in hardware accelerators tailored for efficient DNNs.
- Explored the integration of homomorphic encryption for secure DNN deployment.
Conclusions:
- Efficient DNN deployment requires a multi-faceted approach combining algorithmic compression, specialized hardware, and robust security.
- Further research is needed in hardware evaluation, generalization, and integrating diverse compression strategies.
- The survey provides a roadmap for developing high-performance, cost-efficient, and secure DNNs.
