Tech-invite  3GPPspace  IETFspace
21  22  23  24  25  26  27  28  29  31  32  33  34  35  36  37  38  4‑5x

Content for  TR 22.874  Word version:  18.2.0

Top   Top   Up   Prev   Next
0…   4   5…   5.2…   5.3…   5.4…   5.5…   6…   6.2…   6.3…   6.4…   6.5…   6.6…   6.7…   7…   7.2…   7.3…   7.4…   8…   A…   A.2   A.3   A.4   B   C   D… 
6.7 Prediction of AI/ML model distribution  6.7.1 Description  6.7.2 Pre-conditions  6.7.3a Service Flows for Prediction based on monitoring function of the 5G network  6.7.3b Service Flows for Prediction based on NWDAF analytics function of the 5G network  6.7.4 Post-conditions  6.7.5 Existing features partly or fully covering the use case functionality  6.7.6 Potential Requirements 
...

 

6.7  Prediction of AI/ML model distribution  p. 42

6.7.1  Description  p. 42

If the AI/ML model server was intelligent enough, it would be ideal that the AI/ML model server could predict that the User would need to use some AI/ML model and download the AI/ML model to the UE in advance.
Take image recognition for example, the AI/ML model could be able to predict and download specific AI/ML model in advance based on the user's location, direction of travel and users use of image recognition functionality. For the users located in a famous history museum or moving towards a famous history museum, the 5G network functions and the AI/ML model server could interact with each other and predict that the users could possibly need to use the AI/ML model designed specifically for such history museum which is suitable for the indoor environment and good at historic pictures recognition. Similarly, for the users located in a zoo or moving towards a zoo, the 5G network functions and the AI/ML model server could interact with each other and predict that the users could possibly need to use the AI/ML model designed specifically for animal recognition. In contrast, for the users located in an arboretum or moving towards an arboretum, the 5G network functions and the AI/ML model server could interact with each other and predict that the users could possibly need to use the AI/ML model designed specifically for flower or plant recognition.
The 5G network and the AI/ML model shall not violate the user privacy policies for the prediction or pre-download of AI/ML models.
According to user privacy policies, the AI/ML model server may not to be able to get access to the UE's location information.
If the AI/ML model server is allowed to access to the UE's location information according to user privacy policies, the AI/ML model server could obtain the UE's location information via two options:
Option 1:
the AI/ML model server could obtain the UE's location information directly from the UE, over the application layer. This is pure application layer interaction between the AI/ML based image recognition application and the AI/ML model server.
Option 2:
the AI/ML model server could obtain the UE's location information via the location service provided by the 5G network. This is already specified in TS 23.273 (see further analysis in clause 6.1.4).
If the AI/ML model server is not allowed to access to the UE's location information according to user privacy policies, the 5G network has the UE's location information and is not allowed to expose the UE location information to the AI/ML model server. The 5G network function could still interaction with the AI/ML model server for prediction of need of certain AI/ML model without breaking the user privacy policies. There could be lots of workable solutions. Here daftly describe two possible solutions:
Option a):
Prediction based on monitoring function of the 5G network. New monitoring event could to be introduced to assist the prediction by the AI/ML model server without disclosure of User location information to the AI/ML model server. Option b) Prediction based on NWDAF analytics function of the 5G network. Based on analytics of UE's location, mobility, download data size and etc, the 5G network could predict that lots UE will download certain amount of data from an AI/ML model server in some location area and inform the AI/ML model that certain UE will probably download certain amount of data. The AI/ML model server could use such information to adjust its prediction.
Up

6.7.2  Pre-conditions  p. 43

The UE runs an application providing the capability of AI/ML model inference for image recognition.
An AI/ML server manages the AI/ML model pool, and is capable to download the requested model to the application providing AI/ML based image recognition.
The 5G system is able to provide 5G network related information to the AI/ML server.

6.7.3a  Service Flows for Prediction based on monitoring function of the 5G network  p. 43

  1. The AI/ML model server provides to the 5G network function with some criteria. Each criterion could have different rules. For example, one criterion could contain a list of location areas which contains all the museums, galleries and antique shops all over the country.
  2. If the 5G network functions determine the UE meets one criterion, for example, the UE is located in the list of location areas, then the 5GS network function (e.g., NEF) notify the AI/ML model server which criterion the UE meets.
  3. According the criterion that the UE meets, the AI/ML model server determines that the UE might need to use the AI/ML model designed specifically for such history museum which is suitable for the indoor environment and good at historic pictures recognition.
  4. The 5G operator need to make sure that the criteria provided by the AI/ML model does not break the user privacy policies.
Up

6.7.3b  Service Flows for Prediction based on NWDAF analytics function of the 5G network  p. 43

  1. The NWDAF do the data collection and analytics work for the UE's location, mobility, download data size etc.
  2. Based on the NWDAF analytics, the 5G network could predict that lots UE will download certain amount of data from an AI/ML model server in some location area.
  3. For the AI/ML model server that has subscribed the data analytics service from 5G network, 5G network inform AI/ML model server directly or via NEF that certain UE will probably download certain amount of data.
  4. The AI/ML model server could use such information to adjust its prediction.
Up

6.7.4  Post-conditions  p. 44

The 5G network and the AI/ML model server could predict the need of download certain AI/ML model from the UE and refine the prediction during the interaction.
The 5G network could adjust its resource for download of data from AI/ML without expansion of capacity based on the prediction.
The AI/ML model server could predict the user preference and needs without knowing UE's private information, such as location information.
The UE could get better user experience during the use of AI/ML model application via 5G network.
Up

6.7.5  Existing features partly or fully covering the use case functionality  p. 44

Clause 4.1 of TS 23.273:
Location information for one or multiple target UEs may be requested by and reported to an LCS client or an AF within or external to a PLMN, or a control plane NF within a PLMN. Location information contained in the location request and location information contained in the location response are defined in clause 5.5.
For location request from LCS client (neither in the UE nor in the NG-RAN) or AF external to a PLMN, privacy verification of the target UE shall be enabled to check whether it is allowed to acquire the UE location information based on UE LCS privacy profile and whether the LCS client or the AF is authorised to use the location service as defined in clause 5.4.
Clause 4.15.3.1 of TS 23.502
Currently the Monitoring Events feature defined in TS 23.501 is intended for monitoring of specific events in the 3GPP system and reporting such Monitoring Events via the NEF. Location Reporting is one of the supported monitoring events is specified in Table 4.15.3.1-1 of TS 23.502. However, the current monitoring event features indicates either the Current Location or the Last Known Location of a UE to an AF via NEF, this is not applicable when the AI/ML model server is not allowed to access to the UE's location information according to user privacy policies.
Up

6.7.6  Potential Requirements  p. 44

[P.R.6.7-001]
Subject to user consent, operator policy and regulatory constraints, the 5G system shall support the provision of monitoring information or analytics information to a trusted 3rd party AI/ML server for allowing this 3rd party AI/ML server to make a prediction for a suitable AI/ML model to be downloaded to the concerned UE.

Up   Top   ToC