|
Work item:
|
H.MAS-AP
|
|
Subject/title:
|
Architecture and protocols of multi-algorithm scheduling systems
|
|
Status:
|
[Carried to next study period]
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
-
|
|
Liaison:
|
SG11, ISO/IEC JTC 1/SC 42
|
|
Supporting members:
|
Zhejiang Dahua Technology, Zhejiang Lab, China Mobile, Zhejiang University, China Telecom
|
|
Summary:
|
The demand for computing power in artificial intelligence (AI) multimedia applications continues to increase, the speed of algorithm iteration continues to accelerate, and new algorithm frameworks continue to emerge, resulting in a series of industry pain points, including various types of algorithms, new and old versions of algorithms, and algorithm development standards by various algorithm suppliers are not uniform, multiple algorithm interfaces need to be developed for the same AI multimedia application, resulting in high complexity of algorithm interoperability. Algorithms and computing power resources are strongly coupled, idle computing power resources cannot be shared with other intelligent analysis units, resulting in low resource utilization. When new AI multimedia application scenarios are added, new intelligent analysis units need to be added as well, which greatly increases the cost of users.
The multi-algorithm scheduling (MAS) systems consist of the scheduling unit, the algorithm warehouse unit, the computing power resource management unit, and the algorithm operation units. It provides a unified interface for AI multimedia applications and realizes the management and scheduling of multi-type, multi-version, and multi-framework algorithms from different algorithm suppliers. At the same time, the MAS systems realize the decoupling of algorithms, computing power resources, and AI multimedia applications. After pooling the computing power resources, it can be allocated and recycled flexibly according to the demands of computing power resources by different AI multimedia applications.
This Recommendation specifies the architecture and protocols of the MAS systems for AI-based multimedia applications, which meet the requirements and framework of MAS systems specified in [ITU-T F.MAS]. This Recommendation is intended to guide the design, development, implementation, and application of MAS systems.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2023-08-22 11:06:50
|
|
Last update:
2024-09-20 10:27:07
|
|
