Method for monitoring performance of an artificial intelligence (AI)/machine learning (ML) model or algorithm
Monitoring the performance of an Artificial Intelligence (AI)/Machine Learning (ML) model or algorithm is disclosed herein. In one embodiment, a method performed by a first network node in a radio communication network comprises sending at least one first message to a second network node of the radio communication network, the at least one first message comprising a subscription request to obtain from the second network node one or more historical data elements associated with the AI/ML model or algorithm. At least one second message, comprising one or more historical data elements associated with the AI/ML model or algorithm to be monitored by the first network node, is received from the second network node. At least one performance metric associated with the AI/ML model or algorithm to be monitored is determined based on the one or more historical data elements associated with the AI/ML model.
1 . A method performed by a first network node in a radio communication network to monitor the performance of an Artificial Intelligence, AI,/Machine Learning, ML, model or algorithm, the method comprising:
sending at least one first message to a second network node of the radio communication network, the at least one first message comprising a subscription request to obtain from the second network node one or more historical data elements associated with the AI/ML model or algorithm;
receiving at least one second message from the second network node, the at least one second message comprising one or more historical data elements associated with the AI/ML model or algorithm to be monitored by the first network node; and
determining at least one performance metric associated with the AI/ML model or algorithm to be monitored based on the one or more historical data elements associated with the AI/ML model;
wherein:
the first network node hosts a model training function of the AI/ML model or algorithm to be monitored;
the second network node hosts a data collection function of an AI/ML model or algorithm; and
a third network node hosts a model inference function of the AI/ML model or algorithm.
2 . The method of claim 1 , wherein the subscription request comprises one or more of:
an identifier of the first network node;
an identifier for the subscription request;
an indication of a request type;
an indication of a reason or cause for the request;
one or more indications of a corresponding one or more AI/ML models or algorithms to be monitored;
an indication of a network node hosting a model inference function associated with the AI/ML model or algorithm to be monitored;
one or more indications of a time or period of the collection of data;
an indication of the type of data requested;
a timing-related indications;
one or more filtering criteria indicating a type or scope of requested historical data;
one or more requests to receive from the second network node notifications associated with the historical data;
and one or more conditions pertaining to the sending of historical data.
3 . The method of claim 2 , wherein the indication of the request type comprises an indication to initiate a subscription, an indication to renew a subscription, or an indication to cancel a subscription.
4 . The method of claim 2 , wherein the indication of the type of data requested comprises one or more of:
historical inference input data associated with the AI/ML model to be monitored;
historical inference output associated with the AI/ML model to be monitored, comprising one or more information elements that resulted from each inference step executed by the AI/ML model or algorithm to be monitored; and
historical measurements of information that the AI/ML model or algorithm to be monitored is configured to estimate or predict.
5 . The method of claim 2 , wherein the one or more filtering criteria indicating the type or scope of requested historical data comprises one or more of periods of collection, data selected in a random fashion, data associated with one or more radio network procedure, data related to one or more user equipment, UE, or type of UE, data pertaining to performance indicators, data pertaining to UE or network configuration data, data collected for one or more area of interests, data collected for one or more S-NSSAI, data collected for or one or more 5QI, data collected for one or more services, data collected for MDT, data collected for QoE, radio measurements, load metrics, and data related to energy savings.
6 . The method of claim 2 , wherein the one or more conditions pertaining to the sending of historical data comprises one or more of:
a periodic sending with a reporting periodicity;
a sending based on event;
timing indications;
indications of a size of historical data required to monitor the AI/ML model or algorithm; and
indications to start, stop, pause, or resume sending of historical data.
7 . The method of claim 2 , wherein the one or more requests to receive from the second network node notifications associated with the historical data comprises a request to receive a notification when requested data is available or not available, when historical data of a certain type becomes available, when a modification is occurring or has occurred in the historical data, which type of modification or a quantification of the change, or when sending historical data is expected to start, stop, pause, or resume.
8 . The method of claim 1 , wherein the at least one second message comprises one or more of an identifier of the second network node, an identifier of the subscription request, an identifier of the request type, historical data stored or collected by the second network node associated with the AI/ML model or algorithm to be monitored, and a timing related indications.
9 . The method of claim 1 , wherein:
the at least one second message further comprises an indication of one or more fourth network nodes that can be requested to send historical data; and
wherein the indication comprises one or more of an Internet Protocol, IP, address, a Uniform Resource Locator, URL, and a Uniform Resource Indicator, URI.
10 . The method of claim 1 , wherein the at least one second message further comprises at least one notification providing the first network node with a response to one or more notification requests of the first message.
11 . The method of claim 1 , wherein the performance metric comprises one or more of an indication of accuracy of a model inference function, an indication of whether or not an output of an AI/ML model inference is affected by a certain type of uncertainty, an indication of a type of uncertainty affecting the output of the AI/ML model inference, an indication of a level of uncertainty affecting the output of the AI/ML model inference, and an indication of a bias in an input or in the output of the AI/ML model inference.
12 . The method of claim 1 , further comprising sending a third message to the second network node, the third message comprising a request to subscribe to one or more training data samples associated with the AI/ML model monitored by the first network node.
13 . The method of claim 12 , further comprising receiving a fourth message from the second network node, the fourth message comprising one or more training data samples associated with the AI/ML model or algorithm to be monitored.
14 . A first network node for monitoring the performance of an Artificial Intelligence, AI,/Machine Learning, ML, model or algorithm, the network node comprising:
one or more transmitters;
one or more receivers; and
processing circuitry associated with the one or more transmitters and the one or more receivers, the processing circuitry configured to cause the first network node to:
send at least one first message to a second network node of the radio communication network, the at least one first message comprising a subscription request to obtain from the second network node one or more historical data elements associated with the AI/ML model or algorithm;
receive at least one second message from the second network node, the at least one second message comprising one or more historical data elements associated with the AI/ML model or algorithm to be monitored by the first network node; and
determine at least one performance metric associated with the AI/ML model or algorithm to be monitored based on the one or more historical data elements associated with the AI/ML model;
wherein:
the first network node hosts a model training function of the AI/ML model or algorithm to be monitored;
the second network node hosts a data collection function of an AI/ML model or algorithm; and
a third network node hosts a model inference function of the AI/ML model or algorithm.
15 . A method performed by a second network node in a radio communication network for hosting a data collection function of an Artificial Intelligence, AI,/Machine Learning, ML, model or algorithm, the method comprising:
receiving at least one first message from a first network node of the radio communication network, the at least one first message comprising a subscription request to obtain from the second network node one or more historical data elements associated with the AI/ML model or algorithm;
sending at least one second message to the first network node, the at least one second message comprising one or more historical data elements associated with the AI/ML model or algorithm to be monitored by the first network node;
receiving a third message from the first network node, the third message comprising a request to subscribe to one or more training data samples associated with the AI/ML model monitored by the first network node; and
sending a fourth message to the first network node, the fourth message comprising one or more training data samples associated with the AI/ML model or algorithm to be monitored.
16 . The method of claim 15 , wherein the second message further comprises an indications of one or more fourth network nodes that can be requested to send historical data.
17 . The method of claim 15 , wherein the second message further comprises at least one notification providing the first network node with a response to one or more notification requests of the first message.
18 . A second network node for hosting a data collection function of an Artificial Intelligence, AI,/Machine Learning, ML, model or algorithm, the second network node comprising:
one or more transmitters;
one or more receivers; and
processing circuitry associated with the one or more transmitters and the one or more receivers, the processing circuitry configured to cause the second network node to:
receive at least one first message from a first network node of the radio communication network, the at least one first message comprising a subscription request to obtain from the second network node one or more historical data elements associated with the AI/ML model or algorithm;
send at least one second message to the first network node, the at least one second message comprising one or more historical data elements associated with the AI/ML model or algorithm to be monitored by the first network node;
receive a third message from the first network node, the third message comprising a request to subscribe to one or more training data samples associated with the AI/ML model monitored by the first network node; and
send a fourth message to the first network node, the fourth message comprising one or more training data samples associated with the AI/ML model or algorithm to be monitored.