ITU-T Work Programme

[2025-2028] : [SG21] : [Q5/21]

[Declared patent(s)] - [Associated work]

Work item:

F.AICP-MoE

Subject/title:

General technical requirements and framework for artificial intelligence cloud platform- Mixture of Experts (MoE) capability

Status:

Under study

Approval process:

AAP

Type of work item:

Recommendation

Version:

New

Equivalent number:

Timing:

2027-07 (Medium priority)

Liaison:

ITU-T SG2 WP2, SG13, SG17

Supporting members:

Beijing Baidu Netcom Science Technology Co., Ltd., China Mobile Communications Co. Ltd., China Telecommunications Corporation, China Unicom, Huawei Technologies Co., Ltd., Alibaba China Co. Ltd., Hangzhou Hikvision Digital Technology Co.,Ltd.

Summary:

Mixture of Experts (MoE) is a neural network architecture that dynamically selects a subset of multiple expert networks via a gating mechanism, balancing model capacity and computational efficiency. It can enhance training and inference efficiency while improving scalability. It is inherently suitable for multimodal integration of text, image, and audio data, providing a robust technological foundation for advancing intelligent multimedia applications. This Recommendation specifies the framework and requirements of MoE capability of AI cloud platforms covering model construction, deployment and inference, and service management, and from the perspective of platform functionality provision and internal optimization technologies. It aims to build a comprehensive, user-friendly, and high-performance AI cloud platform supporting the full lifecycle of MoE models, meet the need of both platform users and developers. This Recommendation is applicable to the design, development, evaluation, and selection of AI cloud platforms. It helps industry optimize their platforms and enables more effective, scalable, and high-quality AI-enabled multimedia applications.

Comment:

Reference(s):

[SG21-TD190/WP3 (2025-10)

]