Content for TR 23.700-82 Word version: 19.1.0
5 Key issues
6 Application enablement architecture requirements
7 Application architecture for enabling AI/ML services
8 Solutions
9 Deployment scenarios
10 Business Relationships
11 Overall evaluation
12 Conclusions
$ Change history
5 Key issues p. 16
5.1 Key issue #1: Support of Architecture Enhancement and Functions for Application Layer AI/ML Services p. 16
5.2 Key issue #2: AI/ML-enhanced ADAES p. 17
5.3 Key issue #3: Support for federated learning p. 18
5.4 Key issue #4: Key issue on supporting Vertical FL at enablement layer p. 18
5.5 Key Issue #5: Support for of AI/ML operation splitting between AI/ML endpoints and in-time transfer of AI/ML models p. 19
5.6 Key issue #6: Support for transfer learning p. 20
5.7 Key issue #7: Discovery or Support of Member Selection and Maintenance for Application Layer AIML Service p. 21
6 Application enablement architecture requirements p. 22
6.1 General requirements p. 22
6.2 ADAE capability related requirements p. 22
6.3 AIMLE capability related requirements p. 22
7 Application architecture for enabling AI/ML services p. 22
7.1 General p. 22
7.2 Application enablement architecture p. 23
7.2.1 On-Network AIML Enablement (AIMLE) Functional Architecture p. 23
7.2.2 Off-Network AIML Enablement Functional Architecture p. 25
7.2.3 Architecture representation for supporting ADAE analytics p. 25
7.2.4 Functional Entities Description p. 26
7.2.4.1 General p. 26
7.2.4.2 AIMLE client p. 26
7.2.4.3 AIMLE server p. 26
7.2.4.4 ML repository p. 26
7.2.5 Reference Points Description p. 26
7.2.5.1 General p. 26
7.2.5.2 AIML-UU p. 26
7.2.5.3 AIML-S p. 26
7.2.5.4 AIML-C p. 26
7.2.5.5 AIML-R p. 26
7.2.5.6 AIML-E p. 27
7.2.5.7 ADAE-S p. 27
7.2.5.8 SEAL-X p. 27
8 Solutions p. 27
8.0 Mapping of solutions to key issues p. 27
8.1 Solution #1: ADAE Functional Architecture Enhancement to Support Application Layer AI/ML Services p. 28
8.1.2 Functional Architecture p. 28
8.1.2.1 General p. 28
8.1.2.2 On-Network MTME-enhanced ADAE Functional Architecture p. 28
8.1.2.3 On-Network AIML Enablement Layer Functional Architecture p. 29
8.1.2.4 Coexistence of MTME-enhanced ADAE and AIML enabler p. 29
8.1.3 Functional Entities Description p. 30
8.1.3.1 General p. 30
8.1.3.2 ADAE client p. 30
8.1.3.3 ADAE server p. 30
8.1.3.4 AIML Enablement client p. 30
8.1.3.5 AIML Enablement server p. 30
8.1.4 Reference Points Description p. 30
8.1.4.1 General p. 30
8.1.4.2 ADAE-UU p. 30
8.1.4.3 ADAE-S p. 30
8.1.4.4 ADAE-C p. 31
8.1.4.5 AIML-UU p. 31
8.1.4.6 AIML-S p. 31
8.1.4.7 AIML-C p. 31
8.1.4.8 AIML-E p. 31
8.1.5 Architecture Impacts p. 31
8.1.6 Corresponding APIs p. 31
8.1.7 Solution evaluation p. 31
8.2 Solution #2: ML client information retrieval p. 32
8.2.1 General p. 32
8.2.2 Procedures p. 32
8.2.3 Architecture Impacts p. 32
8.2.4 Corresponding APIs p. 33
8.2.5 Solution evaluation p. 33
8.3 Solution #3: Provision of ML clients to support AI/ML at the application layer p. 33
8.3.1 General p. 33
8.3.2 Procedures p. 34
8.3.3 Architecture Impacts p. 34
8.3.4 Corresponding APIs p. 34
8.3.5 Solution evaluation p. 34
8.4 Solution #4: Support for ML-enabled ADAE analytics p. 35
8.4.1 Solution description p. 35
8.4.2 Architecture Impacts p. 37
8.4.3 Corresponding APIs / information p. 37
8.4.4 Solution evaluation p. 37
8.5 Solution #5: AI/ML model management p. 37
8.5.1 General p. 37
8.5.2 AI/ML model information storage procedure p. 38
8.5.3 AI/ML model information discovery procedure p. 38
8.5.4 Architecture Impacts p. 39
8.5.5 Corresponding APIs p. 40
8.5.5.1 AI/ML model information storage request p. 40
8.5.5.2 AI/ML model information storage response p. 41
8.5.5.3 AI/ML model information discovery request p. 41
8.5.5.4 AI/ML model information discovery response p. 41
8.5.6 Solution evaluation p. 42
8.6 Solution #6: AIML enablement client selection p. 42
8.6.1 Solution description p. 42
8.6.1.1 AIML Enablement Client Selection with SEAL and 5GC support p. 42
8.6.1.2 Dynamic AIML Enablement Client selection and Monitoring Subscription p. 43
8.6.2 Architecture Impacts p. 45
8.6.3 Corresponding APIs p. 45
8.6.4 Solution evaluation p. 46
8.7 Solution #7: AIML enablement client discovery p. 46
8.7.1 Solution description p. 46
8.7.2 Architecture Impacts p. 48
8.7.3 Corresponding APIs p. 48
8.7.4 Solution evaluation p. 49
8.8 Solution #8: AIML enablement client registration p. 49
8.8.1 Solution description p. 49
8.8.2 Architecture Impacts p. 50
8.8.3 Corresponding APIs p. 50
8.8.4 Solution evaluation p. 51
8.9 Solution #9: Support for FL member registration p. 51
8.9.1 Solution description p. 51
8.9.1.1 Procedure for FL member registration p. 51
8.9.1.2 Procedure for FL member registration update p. 52
8.9.2 Architecture Impacts p. 53
8.9.3 Corresponding APIs p. 53
8.9.4 Solution evaluation p. 53
8.10 Solution #10: AI/ML member participation configurations provisioning and management p. 53
8.10.1 General p. 53
8.10.2 Procedures p. 54
8.10.3 Information flows p. 54
8.10.3.1 AI/ML member participation configurations provisioning and management request p. 54
8.10.3.2 AI/ML member participation configurations provisioning and management response p. 55
8.10.4 Architecture Impacts p. 55
8.10.5 Corresponding APIs p. 55
8.10.6 Solution evaluation p. 55
8.11 Solution #11: AIML service lifecycle management procedure p. 56
8.11.1 General p. 56
8.11.2 AIML service lifecycle management procedure p. 56
8.11.3 Architecture Impacts p. 57
8.11.4 Corresponding APIs p. 57
8.11.5 Solution evaluation p. 57
8.12 Solution #12: AI/ML model lifecycle management p. 58
8.12.1 Solution Description on AI/ML Model Lifecycle Management p. 58
8.12.2 Consumer-based ML Model Performance Degradation Detection p. 59
8.12.3 Architecture Impacts p. 60
8.12.4 Corresponding APIs p. 60
8.12.5 Solution evaluation p. 61
8.13 Solution #13: Analytics and Assistance Information Collection for Supporting FL Member (Re)Selection with the ADAES capabilities p. 61
8.13.1 General p. 61
8.13.2 Procedure on Analytics/Assistance Information Collection for FL member (re)selection with the ADAES capabilities p. 61
8.13.3 Architecture Impacts p. 62
8.13.4 Corresponding APIs p. 62
8.13.5 Solution evaluation p. 62
8.14 Solution #14: AI/ML policies provisioning and management p. 62
8.14.1 General p. 62
8.14.2 Procedures p. 63
8.14.3 Information flows p. 63
8.14.3.1 AI/ML service request with policies provisioning and management information p. 63
8.14.3.2 AI/ML service response with policies provisioning and management information p. 66
8.14.4 Architecture Impacts p. 66
8.14.5 Corresponding APIs p. 66
8.14.6 Solution evaluation p. 66
8.15 Solution #15: ADAES support for AI-enabled DN Energy Analytics p. 67
8.15.1 Solution description p. 67
8.15.2 Architecture Impacts p. 68
8.15.3 Corresponding APIs p. 68
8.15.4 Solution evaluation p. 68
8.16 Solution #16: Support for FL event notifications p. 69
8.16.1 Solution description p. 69
8.16.1.1 Procedure on subscription for FL related events p. 69
8.16.1.2 Procedure on FL related event notification p. 69
8.16.1.3 List of FL-related Events p. 70
8.16.2 Architecture Impacts p. 71
8.16.3 Corresponding APIs p. 71
8.16.4 Solution evaluation p. 71
8.17 Solution #17: AIML operational management p. 72
8.17.1 Solution description p. 72
8.17.1.1 AIML operational management procedure p. 72
8.17.1.2 AIML enablement client task triggering p. 74
8.17.2 Architecture Impacts p. 75
8.17.3 Corresponding AIML-S APIs p. 75
8.17.5 Solution evaluation p. 78
8.18 Solution #18: Supporting VFL in Enablement Layer p. 78
8.18.1 Solution description p. 78
8.18.2 Architecture Impacts p. 80
8.18.3 Corresponding APIs p. 80
8.18.4 Solution evaluation p. 80
8.19 Solution #19: Support for AIML operation splitting p. 81
8.19.1 Solution description p. 81
8.19.2 Procedure for AIML operation splitting discovery p. 81
8.19.2.1 Procedure for subscribe-notify for split operation pipeline events p. 83
8.19.2.2 Procedure for VAL server AIML split operation registration p. 83
8.19.2.3 Procedure for client creating pipeline p. 84
8.19.3 Architecture Impacts p. 84
8.19.4 Corresponding APIs p. 85
8.19.4.1 API Overview p. 85
8.19.4.2 Information flows p. 85
8.19.5 Solution evaluation p. 88
8.20 Solution #20: AIML Enabler support for Transfer Learning p. 88
8.20.1 Solution description p. 88
8.20.2 Architecture Impacts p. 89
8.20.3 Corresponding APIs p. 90
8.20.4 Solution evaluation p. 90
8.21 Solution #21: AIML data management procedure p. 90
8.21.1 Solution description p. 90
8.21.2 Architecture Impacts p. 92
8.21.3 Corresponding APIs p. 92
8.21.4 Solution evaluation p. 93
8.22 Solution #22: Horizontal Federated Learning training p. 93
8.22.1 Solution description p. 93
8.22.2 Architecture Impacts p. 95
8.22.3 Corresponding APIs p. 95
8.22.4 Solution evaluation p. 95
8.23 Solution #23: AIML services in edge p. 96
8.23.1 Solution description p. 96
8.23.2 Architecture Impacts p. 98
8.23.3 Corresponding AIML-S APIs p. 98
8.23.4 Corresponding AIML-UU APIs p. 99
8.23.5 Solution evaluation p. 99
8.24 Solution #24: Dynamic ML model distribution p. 99
8.24.1 Solution description p. 99
8.24.2 Architecture Impacts p. 101
8.24.3 Corresponding APIs p. 101
8.24.4 Solution evaluation p. 103
8.25 Solution #25: Support for AIML model distribution p. 103
8.25.1 Solution description p. 103
8.25.2 Procedures p. 103
8.25.2.1 Subscription for AIML model updates p. 103
8.25.2.2 Notification for AIML model updates p. 105
8.25.3 Architecture Impacts p. 106
8.25.4 Corresponding APIs p. 106
8.25.4.1 API Overview p. 106
8.25.5 Solution evaluation p. 106
8.26 Solution #26: Support for FL member grouping p. 106
8.26.1 Solution description p. 106
8.26.2 Architecture Impacts p. 107
8.26.3 Corresponding APIs p. 108
8.26.4 Solution evaluation p. 108
8.27 Solution #27: New ADAE Analytics for Supporting FL Member (re-) selection p. 108
8.27.1 General p. 108
8.27.2 Procedure p. 109
8.27.3 Information flows p. 110
8.27.3.1 General p. 110
8.27.3.2 Application Layer AI/ML Member capability analytics subscription request p. 110
8.27.3.3 Application Layer AI/ML Member capability analytics subscription response p. 111
8.27.3.4 Application layer AI/ML Member capability analytics notification p. 111
8.27.3.5 Application Layer AI/ML Member capability data collection subscription request p. 112
8.27.3.6 Application Layer AI/ML Member capability data collection subscription response p. 112
8.27.3.7 Data Notification p. 112
8.27.3.8 Get analytics data request p. 113
8.27.3.9 Get analytics data response p. 113
8.27.4 Architecture Impacts p. 114
8.27.5 Corresponding APIs p. 114
8.15.6 Solution evaluation p. 114
8.28 Solution #28: Support AI/ML Splitting Operations in Enablement Layer p. 114
8.28.1 General p. 114
8.28.2 Procedures for supporting assistance of split AI/ML operations p. 115
8.28.2.1 Procedure for subscribe/request assistance of split AI/ML operations p. 115
8.28.2.2 Procedure for assistance of AI/ML task/model/data delivery/distribution p. 116
8.28.3 Architecture Impacts p. 117
8.28.4 Corresponding APIs p. 117
8.28.5 Solution evaluation p. 117
8.29 Solution #29: Enhance AIML Services for Assisting Edge Computing p. 118
8.29.1 General p. 118
8.29.2 Procedures for assisting edge computing p. 118
8.29.3 Architecture Impacts p. 119
8.29.4 Corresponding APIs p. 119
8.29.5 Solution evaluation p. 119
8.30 Solution #30: Support Transfer of Intermediate AIML Operation Information p. 119
8.30.1 General p. 119
8.30.2 Procedures for intermediate AI/ML information transfer p. 120
8.30.2.1 AI/ML Enablement Server assist intermediate AI/ML information transfer p. 120
8.30.2.2 Direct intermediate AI/ML information transfer p. 121
8.30.3 Architecture Impacts p. 122
8.30.4 Corresponding APIs p. 122
8.30.5 Solution evaluation p. 123
8.31 Solution #31: Supporting AIML inference service in edge p. 123
8.31.1 Solution description p. 123
8.31.2 Architecture Impacts p. 124
8.31.3 Corresponding APIs p. 124
8.31.4 Solution evaluation p. 124
9 Deployment scenarios p. 125
9.1 General p. 125
9.2 Deployment model #1: Cloud-deployed AIMLE server p. 125
9.3 Deployment model #2 Edge-deployed AIMLE server p. 125
9.4 Deployment model #3: Hierarchical AIMLE server deployment p. 126
10 Business Relationships p. 127
11 Overall evaluation p. 127
11.1 Summary of enablement capabilities and APIs p. 127
11.2 Key issue #1: Support of Architecture Enhancement and Functions for Application Layer AI/ML Services p. 129
11.3 Key issue #2: AI/ML-enhanced ADAES p. 130
11.4 Key issue #3: Support for federated learning p. 131
11.5 Key issue #4: Supporting Vertical FL at enablement layer p. 131
11.6 Key Issue #5: Support for of AI/ML operation splitting between AI/ML endpoints and in-time transfer of AI/ML models p. 132
11.6.1 Split AI/ML operation p. 132
11.6.2 Transfer of AI/ML models p. 133
11.7 Key issue #6: Support for transfer learning p. 133
11.8 Key issue #7: Discovery or Support of Member Selection and Maintenance for Application Layer AIML Service p. 133
12 Conclusions p. 134
12.0 General conclusions and recommendations p. 134
12.1 Conclusions of key issue #1 p. 135
12.2 Conclusions of key issue #2 p. 135
12.3 Conclusions of key issue #3 p. 135
12.4 Conclusions of key issue #4 p. 136
12.5 Conclusions of Key Issue #5 p. 136
12.6 Conclusions of key issue #6 p. 137
12.7 Conclusions of key issue #7 p. 137
$ Change history p. 138
