Training Device Discovery for NWDAF Inference Addressing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The deployment of training and inference modules in a network data analytics function (NWDAF) results in high costs due to the separate deployment of the training device in a centralized, cloud-based manner and the inference device in a distributed manner, necessitating a method to correctly address the training device from the inference device.
Innovation Solution
An inference device sends a request to a service discovery entity for information about training devices, including algorithm type or identifier, and receives capability information to select an appropriate training device based on preset conditions, ensuring smooth model training and reducing deployment costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If the training module and inference module are separately deployed as training device and inference device, then deployment costs are reduced, but the inference device cannot correctly address the training device
Solution Approach 1:
A service discovery entity is introduced as an intermediary between the inference device and training device. The inference device sends a service discovery request to the service discovery entity, which returns service discovery information containing the address of a suitable training device. This mediator resolves the addressing problem without requiring direct knowledge of training device locations at the inference device.
Solution Approach 2:
The service discovery entity pre-stores information about multiple training devices including their addresses and capability information. When the inference device needs training services, the addressing information is already available in the service discovery entity, eliminating the need for real-time discovery or hardcoding of training device addresses.
2Adaptability or versatility
If multiple training devices are deployed with different locations, load, and supported AI capabilities, then service flexibility is improved, but device selection complexity increases
Solution Approach 1:
The service discovery entity provides feedback to the inference device by returning service discovery information that includes capability information of training devices (such as supported AI capabilities and current load). This feedback mechanism enables the inference device to make informed selection decisions based on actual training device states rather than attempting to manage complexity directly.
Solution Approach 2:
The service discovery entity serves multiple functions: it stores training device information, matches inference device requirements with training device capabilities, and provides addressing information. This multi-functional component handles the complexity of device selection centrally, allowing individual devices to remain simple while the system maintains high flexibility.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A method and an apparatus for determining device information and a system are provided, and relate to the field of communications technologies, so that an inference device can correctly address a training device. The method includes: An inference device sends a first request to a service discovery entity. The first request is used to request information about one or more training devices, and the first request includes an algorithm type or an algorithm identifier of a first model requested by the inference device. The inference device receives the information about the one or more training devices from the service discovery entity. The information about the training device includes capability information. The inference device determines a first training device from the information about the one or more training devices based on a preset condition.