|
Work item:
|
F.AI-AKS
|
|
Subject/title:
|
Reference model of AI knowledge sharing for AI computing platform
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2025-Q2 (Medium priority)
|
|
Liaison:
|
ITU-T SG13
|
|
Supporting members:
|
ETRI, Hankuk University, Rep.of Korea
|
|
Summary:
|
Artificial intelligence (AI) application services have gained substantial recognition due to their remarkable capacity to efficiently harness the potential of Massive IoT Devices (MID) in a wide array of domains. For the effective development of these AI application services, it is necessary to consider the various computing powers of MIDs because they encompass a broad spectrum of types, ranging from low-performance edge devices to high-performance edge devices. As an example, AI services that require significant computational power are typically executed on high-performance edge devices. Nonetheless, when executing these services on a low-performance edge device, they may encounter issues such as sluggish performance or even complete halts. Consequently, loading and running services that involve extensive computation and complex technologies like AI on low-performance edge devices is challenging. To tackle this challenge, efforts are underway to address the issue through optimization techniques (e.g, AI network structure, convolution) or lightweight solutions (e.g, weight/channel pruning, quantization). However, resolving all these challenges solely through optimization or lightweight methods is not a straightforward task due to inherent differences in device characteristics, AI service specifications, user requirements, and performance expectations.
One effective approach to tackle these challenges is to provide a device that satisfies the user's requirements or adjust the functionality of the program according to the requirements. By ensuring that the device meets the user's specifications or making necessary adaptations to the program, it becomes feasible to overcome the limitations posed by low-performance edge devices and enhance the overall user experience. For this, an AI computing platform becomes essential. This platform plays a crucial role in enhancing both hardware performance and software functionality based on diverse needs, including user requirements, service requirements, and performance requirements. The primary objective of this platform is to empower users in intelligently managing and controlling the physical world by augmenting resources or enhancing program functions to meet specific requirements.
This Recommendation specifies an AI knowledge sharing reference model that provides smooth functionality and ensures optimal performance according to various requirements. This reference model serves as a comprehensive framework that captures the various requirements essential for the execution of AI services. It provides valuable information such as learning data and learning models that are indispensable during the development of AI services. By leveraging this knowledge sharing reference model, developers and users are provided with an execution environment to effectively build and deploy AI services. Ultimately, the knowledge sharing reference model facilitates the seamless execution of AI services and promotes collaboration and innovation within the AI ecosystem.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2023-08-22 10:07:04
|
|
Last update:
2023-12-18 10:56:16
|
|
