边缘内生智能白皮书-英文_124页_4mb

Edge-Native Intelligence: Combining Edge Computing and AI for 6G

Definition and Importance

Edge-Native Intelligence integrates edge computing and AI to enhance wireless communication systems by reducing latency, improving data processing at the network edge, and supporting new scenarios like 6G. It enables decentralized intelligence deployment, addressing challenges such as high data volumes, low latency requirements, and security risks. This approach improves user service experiences and supports applications in areas like autonomous driving and smart cities.

6G Edge Intelligence Architecture

The architecture is based on "four layers" (infrastructure, virtualization, function, application) and "three planes" (control, AI, management). It incorporates microservices, container orchestration, and decoupled AI functions. Key components include edge servers, intelligent gateways, and cloud-edge collaboration mechanisms to provide customized AI services and efficient resource management.

Key Technologies

• Model Lightweighting: Techniques like pruning, distillation, quantization, and Neural Architecture Search (NAS) reduce model size and computational load for efficient edge deployment.
• Edge-Cloud Collaborative Intelligence: Includes Federated Learning for privacy-preserving distributed training, Split Learning for reduced communication overhead, and Model Partition for balancing inference accuracy and latency.
• Wireless Federated Learning: Optimizes communication and resource scheduling in edge environments, enabling collaborative AI training at the network edge while addressing data heterogeneity and security challenges.

Applications

• Smart Transportation: Enables real-time traffic monitoring, event detection, and autonomous vehicle coordination through edge-AI collaboration.
• Smart Manufacturing: Supports predictive maintenance, quality control, and flexible production scheduling by leveraging edge devices for low-latency AI tasks.
• Intelligent Energy Saving: Optimizes network energy consumption and base station operations to reduce operational costs and enhance sustainability.

Challenges

• High communication overhead in wireless environments, especially with iterative model training in Federated Learning.
• Limited resource constraints on edge devices, requiring efficient hardware and software optimizations.
• Security and privacy issues, such as data protection and secure collaboration in edge intelligence deployment.