Machine learning configuration information transfer to UE using proximity services (ProSe)/sidelink wireless communication

A first user device comprising:

• at least one processor;
• and
• at least one memory including computer program code;
• the at least one memory and the computer program code configured to, with the at least one processor, cause the first user device at least to:
• receive, from a network node, a first indication that the first user device is selected to act as a user device-network relay to relay Machine Learning (ML) configuration information, wherein the network node comprises Proximity Services (ProSe) application server, and wherein the first indication comprises a ProSe ML application identifier;
• in response to receiving the first indication:
• activate an ML configuration support;
• and
• perform a sidelink discovery procedure with a plurality of user devices in proximity to the first user device for determining at least one second user device from the plurality of user devices that:
• is capable of sidelink communications,
• utilizes an ML model to perform one or more ML-related operations, and
• is capable of performing a ML-enabled function for which the first user device transmits the ML configuration information;
• transmit, to the at least one second user device, a notification indicating that the first user device has a capability to provide the ML configuration information via sidelink communications, the transmitting of the notification comprising at least one of:
• transmitting an announcement message, wherein the announcement message comprises a user device to network relay discovery announcement message;
• or
• receiving, from the at least one second user device, a solicitation message requesting information about relay capabilities of the first user device, and in response to the solicitation message, transmitting, to the at least one second user device, a response message indicating the capability of the first user device to provide the ML configuration information;
• report, to the network node, an identification of one or more target devices including the at least one second user device capable to receive the relay ML configuration information;
• receive, from the network node, the ML configuration information and a network-configured threshold value, wherein the ML configuration information comprises:
• a plurality of parameters to configure the ML model of the at least one second user device, the plurality of parameters comprising:
• a plurality of weights,
• a plurality of bias values,
• a plurality of activation functions,
• a plurality of convolution parameters,
• a plurality of dropout rates, and
• a plurality of hyperparameters;
• configuration parameters for input data pre-processing and output inference post-processing;
• a configuration for a non-machine learning fallback algorithm;
• one or more timers, including an initial value for at least one of the one or more timers, for determining a state of the at least one second user device as one of:
• served by a network node, or
• not served by the network node;
• a first set of ML-enabled function parameters instructing operation in a first indicated ML operational mode when in the state of being served by the network node;
• and
• a second set of ML-enabled function parameters instructing operation in a second indicated ML operational mode when in the state of not being served by the network node, wherein the second indicated ML operational mode is different from the first indicated ML operational mode;
• receive, from the network node, a second indication to modify the ML configuration information;
• in response to receiving the second indication, modify the received ML configuration information to generate modified ML configuration information, the modifying comprising:
• modifying one or more of the plurality of weights or the plurality of hyperparameters of the ML model;
• modifying the input data pre-processing or the output inference post-processing configuration parameters,
• modifying a fallback configuration,
• adjusting a validity time period of the ML configuration information, and
• adjusting a complexity of the ML configuration information by removing one or more layers of the ML model or decreasing an input size of the input data for the ML model;
• and
• transmit, to the at least one second user device via sidelink communications, the modified ML configuration information for carrying out ML-related operation, the transmit comprising:
• determining a user device density of the one or more target devices;
• comparing the determined user device density to the network-configured threshold value to generate a comparison result;
• in response to the comparison result indicating the determined user device density is less than the network-configured threshold value, transmitting the modified ML configuration information to the at least one second user device via unicast communications;
• and
• in response to the comparison result indicating the determined user device density is greater than the network-configured threshold value, transmitting the modified ML configuration information to the one or more target devices via groupcast or broadcast communications;
• receive, from the at least one second user device subsequent to the transmitting of the modified ML configuration information, a status notification indicating which of the first indicated ML operational mode or the second indicated ML operational mode is being used by the at least one second user device;
• and
• in response to receiving the status notification, relay the status notification to the network node.

A method performed by a second user device, the method comprising:

• receiving, from a first user device, a notification that the first user device has a capability to provide machine learning (ML) configuration information via sidelink communications, wherein the second user device utilizes an ML model to perform one or more ML-related operations, and wherein the notification is received by performing:
• receiving, from the first user device, an announcement message comprising the notification;
• and
• transmitting, to the first user device via sidelink communications, a solicitation message requesting information about relay capabilities of the first user device, and in response to the solicitation message, receiving, from the first user device, a response message comprising the notification;
• receiving, from the first user device, the ML configuration information via one of unicast communications or groupcast or broadcast communications,
• wherein the ML configuration information has been modified by the first user device before being received by the second user device, and
• wherein the ML configuration information comprises:
• a plurality of parameters to configure the ML model of the second user device, the plurality of parameters comprising:
• a plurality of weights,
• a plurality of bias values,
• a plurality of activation functions,
• a plurality of convolution parameters,
• a plurality of dropout rates, and
• a plurality of hyperparameters;
• configuration parameters for input data pre-processing and output inference post-processing;
• a configuration for a non-machine learning fallback algorithm;
• one or more timers, including an initial value for at least one of the one or more timers, for determining a state of the second user device as one of:
• served by a network node, or
• not served by the network node;
• a first set of ML-enabled function parameters instructing operation in a first indicated ML operational mode when in the state of being served by the network node;
• and
• a second set of ML-enabled function parameters instructing operation in a second indicated ML operational mode when in the state of not being served by the network node, wherein the second indicated ML operational mode is different from the first indicated ML operational mode;
• using, by the second user device, the received ML configuration information to configure and operate the ML model at the second user device to carry out an ML-related operation, wherein using the received ML configuration comprises:
• based on the one or more timers, determining the state of the second user device as either served by the network node or not served by the network node;
• and
• selectively operating the ML model in one of:
• a first mode when the second user device is in the state of being served by the network node by using the first set of ML-enabled function parameters;
• or
• a second mode different from the first mode when the second user device is in the state of not being served by the network node by using the second set of ML-enabled function parameters, wherein the first mode and the second mode are selected from a group consisting of an inference mode, a training mode, and an off mode;
• and
• sending, to the network node via the first user device, a notification indicating the operating mode of the ML model of the second user device.

A second user device comprising:

• at least one processor;
• and
• at least one memory including computer program code;
• the at least one memory and the computer program code configured to, with the at least one processor, cause the second user device at least to:
• receive, from a first user device, a notification that the first user device has a capability to provide machine learning (ML) configuration information via sidelink communications, wherein the second user device utilizes an ML model to perform one or more ML-related operations, and wherein the notification is received by performing:
• receiving, from the first user device, an announcement message comprising the notification;
• or
• transmitting, to the first user device via sidelink communications, a solicitation message requesting information about relay capabilities of the first user device, and in response to the solicitation message, receive, from the first user device, a response message comprising the notification;
• receive, from the first user device, the ML configuration information via one of unicast communications or groupcast or broadcast communications,
• wherein the ML configuration information has been modified by the first user device before being received by the second user device, and
• wherein the ML configuration information comprises:
• a plurality of parameters to configure the ML model of the second user device, the plurality of parameters comprising:
• a plurality of weights,
• a plurality of bias values,
• a plurality of activation functions,
• a plurality of convolution parameters,
• a plurality of dropout rates, and
• a plurality of hyperparameters;
• configuration parameters for input data pre-processing and output inference post-processing;
• a configuration for a non-machine learning fallback algorithm;
• one or more timers, including an initial value for at least one of the one or more timers, for determining a state of the second user device as one of:
• served by a network node, or
• not served by the network node;
• a first set of ML-enabled function parameters instructing operation in a first indicated ML operational mode when in the state of being served by the network node;
• and
• a second set of ML-enabled function parameters instructing operation in a second indicated ML operational mode when in the state of not being served by the network node, wherein the second indicated ML operational mode is different from the first indicated ML operational mode;
• use, by the second user device, the received ML configuration information to configure and operate the ML model at the second user device to carry out an ML-related operation, wherein using the received ML configuration comprises:
• based on the one or more timers, determining the state of the second user device as either served by the network node or not served by the network node;
• and
• selectively operating the ML model in one of:
• a first mode when the second user device is in the state of being served by the network node by using the first set of ML-enabled function parameters;
• or
• a second mode different from the first mode when the second user device is in the state of not being served by the network node by using the second set of ML-enabled function parameters, wherein the first mode and the second mode are selected from a group consisting of an inference mode, a training mode, and an off mode;
• and
• sending, to the network node via the first user device, a notification indicating the operating mode of the ML model of the second user device.