Content for TR 22.876 Word version: 19.1.0
0 Introduction
1 Scope
2 References
3 Definitions, symbols and abbreviations
3.1 Definitions
3.2 Symbols
3.3 Abbreviations
4 Overview
...
...
0 Introduction p. 5
This document identifies use cases with functional and performance requirements for 5G system support of AI/ML training and inference using direct device connection. By using direct device connection, it can provide a communication capacity and coverage, more work task offloading choice to achieve an efficient AIML training and inference.
1 Scope p. 6
The objective of this document is to study the use cases with potential functional and performance requirements to support efficient AI/ML operations using direct device connection for various applications e.g. auto-driving, robot remote control, video recognition, etc.
The aspects addressed in the document includes:
- Identify the use cases for distributed AI inference;
- Identify the use cases for distributed/decentralized model training;
- Gap analysis to existing 5GS mechanism to support the distributed AI inference and model training.
2 References p. 6
The following documents contain provisions which, through reference in this text, constitute provisions of the present document.
- References are either specific (identified by date of publication, edition number, version number, etc.) or non-specific.
- For a specific reference, subsequent revisions do not apply.
- For a non-specific reference, the latest version applies. In the case of a reference to a 3GPP document (including a GSM document), a non-specific reference implicitly refers to the latest version of that document in the same Release as the present document.
[1]
TR 21.905: "Vocabulary for 3GPP Specifications".
[2]
TR 22.874: Study on traffic characteristics and performance requirements for AI/ML model transfer in 5GS (Release 18).
[3]
TS 22.104: Service requirements for cyber-physical control applications in vertical domains.
[4]
Huaijiang Zhu, Manali Sharma, Kai Pfeiffer, Marco Mezzavilla, Jia Shen, Sundeep Rangan, and Ludovic Righetti, "Enabling Remote Whole-body Control with 5G Edge Computing", to appear, in Proc. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems. Available at: https://arxiv.org/pdf/2008.08243.pdf
[5]
B. Kehoe, S. Patil, P. Abbeel, and K. Goldberg, "A survey of research on cloud robotics and automation," IEEE Transactions on automation science and engineering, vol. 12, no. 2, pp. 398-409, 2015.
[6]
M. Chen, K. Huang, W. Saad, M. Bennis, A. V. Feljan, and H. V. Poor, "Distributed Learning in Wireless Networks: Recent Progress and Future Challenges"IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 39, NO. 12, DECEMBER 2021
[7]
M. Chen, H. V. Poor, W. Saad, and S. Cui, "Wireless communications for collaborative federated learning," IEEE Commun. Mag., vol. 58, no. 12, pp. 48-54, Dec. 2020
[8]
Jianmin Chen, Xinghao Pan, Rajat Monga, Samy Bengio, Rafal Jozefowicz, "Revisiting Distributed Synchronous SGD," arXiv preprint arXiv: 1604.00981, 2016
[9]
Shuxin Zheng, Qi Meng, Taifang Wang, Wei Chen, Nenghai Yu, Zhi-Ming Ma, Tie-Yan Liu, "Asynchronous Stochastic Gradient Descent with Delay Compression" arXiv: 1609.08326, 2020
[10]
TR 21.905: "Vocabulary for 3GPP Specifications".
[11]
Yusuf Aytar, Carl Vondrick, Antonio Torralba: "SoundNet: Learning Sound Representations from Unlabeled Video", 27 Oct 2016.
[12]
Iacovos Ioannou et al.: "Distributed Artificial Intelligence Solution for D2D Communication in 5G Networks", 20 Jan 2020.
[13]
Pimmy Gandotra et al.: "Device-to-Device Communication in Cellular Networks: A Survey".
[14]
Davide Villa et al.: "Internet of Robotic Things: Current Technologies, Applications, Challenges and Future Directions", 15 Jan 2021.
[15]
Charles R. Qi et al.: "PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation", 10 Apr 2017.
[16]
TS 22.261: "Service requirements for the 5G system".
[17]
TS 23.303: "Proximity-based services (ProSe); Stage 2".
[18]
TS 22.104: "Service requirements for cyber-physical control applications in vertical domains; Stage 1".
[19]
[20]
Y. Kang et al., "Neurosurgeon: Collaborative intelligence between the cloud and mobile edge", ACM SIGPLAN Notices, vol. 52, no. 4, pp. 615-629, 2017.
[21]
A. Krizhevsky, I. Sutskever, and G. E. Hinton, "ImageNet classification with deep convolutional neural networks", in Proc. NIPS, 2012, pp. 1097-1105.
[22]
K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition", in Proc. IEEE CVPR, Jun. 2016, pp. 770-778.
[23]
Zhang Z, Wang S, Hong Y, et al. Distributed dynamic map fusion via federated learning for intelligent networked vehicles[C]//2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021: 953-959.
[24]
[25]
To Transfer or Not To Transfer, Massachusetts Institute of Technology, MIT, Michael T. Rosenstein, et al.
[26]
Wang, J. et al. Easy Transfer Learning by Exploiting Intra-domain Structures. In 2019 IEEE International Conference on Multimedia and Expo (ICME), pages 1210-1215 IEEE.
[27]
Wang K C, Fu Y, Li K, et al. Variational model inversion attacks[J]. Advances in Neural Information Processing Systems, 2021, 34: 9706-9719.
[28]
Ming-Fang Chang, John Lambert, Patsorn Sangkloy, et. al. Argoverse: 3D Tracking and Forecasting with Rich Maps. arXiv:1911.02620v1 [cs.CV] 6 Nov 2019.
3 Definitions, symbols and abbreviations p. 7
3.1 Definitions p. 7
For the purposes of the present document, the terms and definitions given in TR 21.905 and the following apply. A term defined in the present document takes precedence over the definition of the same term, if any, in TR 21.905.
Proximity-based work task offloading:
based on 3rd party's request, a relay UE receives data from a remote UE via direct device connection and performs calculation of a work task for the remote UE. The calculation result can be further sent to network server.
3.2 Symbols p. 8
For the purposes of the present document, the following symbols apply:
3.3 Abbreviations p. 8
4 Overview p. 8
In TR 22.874, three types of AIML operations as below has been described:
- AI/ML operation splitting between AI/ML endpoints;
- AI/ML model/data distribution and sharing over 5G system;
- Distributed/Federated Learning over 5G system.
