Abstract: This application provides a method and a related communication apparatus applicable to the communication field. In the method, a network device receives a first notification message sent by a terminal device, where the first notification message indicates that the terminal device has completed training on a target model, and the first notification message includes a model name corresponding to the target model. The network device then determines policy information based on the uplink channel state information of the terminal device; and sends a lightweight indication message to the terminal device.

Abstract: This application provides a model data sending method and an apparatus. The method can reduce an accuracy loss of model data, and includes: A first device determines second information based on first information, where the second information is used by a second device to quantize first model data, the first information includes an evaluation loss corresponding to a current round of training, the second information includes a quantization error threshold, and the first model data is model data that is after the current round of training; the first device sends the second information to the second device; and the first device receives a first message sent by the second device, where the first message includes quantized first model data and first quantization configuration information.

Abstract: A federated learning method, apparatus, and system are disclosed. A first node obtains data distribution information of a plurality of second nodes based on a target data feature required by a training task; the first node selects at least two target second nodes from the plurality of second nodes based on a target data class required by the training task and the data distribution information of the plurality of second nodes; and the first node indicates the at least two target second nodes to perform federated learning, to obtain a federated learning model that is in the training task and that corresponds to the target data class. In this way, when participants have a plurality of data distributions, a trained model is prevented, as much as possible, from being affected by data poisoning.

Abstract: Embodiments of this application provide a communication method, apparatus. One example method includes: a first data analytics network element receive a first request from a sec data analytics network element, wherein the first request carries an analytics type identifier and second model requirement information, and the first request requests information of a model that corresponds to the analytics type identifier and meets the second model requirement information. The second data analytics network element receives the information of the model from the first data analytics network element.