Information processing method, device, equipment and storage medium

Abstract

The present invention discloses an information processing method, device, equipment and storage medium. The method includes: receiving first information respectively sent by at least two second nodes; the first information represents the relevant parameters of the machine learning model trained by the second node; for each second node, based on the first information, judging whether the performance of the machine learning model trained by the corresponding second node meets the preset conditions; when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset conditions, determining a first learning mode corresponding to the corresponding second node, and notifying the corresponding second node of the first learning mode; wherein the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets the preset conditions.

Description

Technical Field

[0001] The present invention relates to the field of wireless network technology, and in particular to an information processing method, device, equipment and storage medium. Background Art

[0002] With the rapid development of communication network technology and artificial intelligence technology, machine learning models can be trained on network element devices, and the trained machine learning models can be used to predict the performance indicators of stored network data. Usually, in the daily maintenance of network element devices, such as cutting and restarting, the machine learning models trained on the network element devices may become inapplicable, resulting in a decrease in the accuracy of the performance indicators of the network data predicted by the network element devices.

[0003] Therefore, it is urgent to find a technical solution to improve the accuracy of performance indicators of network data predicted by network element devices. Summary of the invention

[0004] In view of this, embodiments of the present invention are intended to provide an information processing method, apparatus, device, and storage medium.

[0005] The technical solution of the embodiment of the present invention is achieved as follows:

[0006] At least one embodiment of the present invention provides an information processing method, the method comprising:

[0007] Receiving first information respectively sent by at least two second nodes; the first information represents relevant parameters of the machine learning model trained by the second node;

[0008] For each second node, based on the first information, determining whether the performance of the machine learning model trained by the corresponding second node meets a preset condition;

[0009] When it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset condition, determining a first learning mode corresponding to the corresponding second node, and notifying the corresponding second node of the first learning mode;

[0010] Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0011] In addition, according to at least one embodiment of the present invention, the first information includes a second learning mode of the second node training the machine learning model; and determining the first learning mode corresponding to the corresponding second node includes:

[0012] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode;

[0013] When it is determined that the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain a plurality of nodes to be processed;

[0014] Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result;

[0015] Determine the machine learning model with the best performance evaluation in the evaluation results; use the transfer learning mode corresponding to the machine learning model with the best performance evaluation as the first learning mode corresponding to the corresponding second node;

[0016] The transfer learning mode characterizes that the parameters of the machine learning model are obtained through transfer learning.

[0017] In addition, according to at least one embodiment of the present invention, the notifying the corresponding second node of the first learning mode includes:

[0018] Generate second information; the second information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model;

[0019] The second information is sent to the corresponding second node.

[0020] In addition, according to at least one embodiment of the present invention, the first information includes a second learning mode of the second node training the machine learning model; and determining the first learning mode corresponding to the corresponding second node includes:

[0021] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning modes of the machine learning models trained by the other nodes are not transfer learning modes;

[0022] When it is determined that the learning modes of the machine learning models trained by the other nodes are not transfer learning modes, the autonomous learning mode is used as the first learning mode corresponding to the corresponding second node;

[0023] Among them, the autonomous learning mode represents that the corresponding second node keeps the model parameters unchanged when retraining the machine learning model.

[0024] In addition, according to at least one embodiment of the present invention, the notifying the corresponding second node of the first learning mode includes:

[0025] generating third information; the third information including the name or number of the corresponding second node and the recommended first learning mode;

[0026] The third information is sent to the corresponding second node.

[0027] In addition, according to at least one embodiment of the present invention, the first information includes a second learning mode of the second node training the machine learning model; and determining the first learning mode corresponding to the corresponding second node includes:

[0028] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the learning mode of the machine learning model trained by the other nodes is a transfer learning mode;

[0029] When it is determined that the learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain multiple nodes to be processed;

[0030] Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result;

[0031] The combination of the transfer learning mode and the autonomous learning mode corresponding to the machine learning model whose performance evaluation in the evaluation result is greater than the performance threshold is used as the first learning mode corresponding to the corresponding second node;

[0032] Among them, the combination of the transfer learning mode and the autonomous learning mode represents that the corresponding second node retrains the machine learning model while keeping the model parameters unchanged and while updating the model parameters.

[0033] In addition, according to at least one embodiment of the present invention, the sending to the corresponding second node includes:

[0034] Generate fourth information; the fourth information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model;

[0035] The fourth information is sent to the corresponding second node.

[0036] At least one embodiment of the present invention provides an information processing method, which is applied to a second node. The method includes:

[0037] Infer the predicted value output by the trained machine learning model to obtain the inference result;

[0038] Evaluate the inference result to obtain an evaluation result; determine whether to report the evaluation result to the first node; when it is determined to report the evaluation result to the first node, generate first information; the first information represents relevant parameters of the machine learning model trained by the second node; and send the first information to the first node.

[0039] In addition, according to at least one embodiment of the present invention, generating the first information includes:

[0040] The first information is generated based on the name or number of the second node, and the learning mode and parameters of the machine learning model trained by the second node.

[0041] In addition, according to at least one embodiment of the present invention, the method further includes:

[0042] Receive a first learning mode sent by the first node; the first learning mode is determined by the first node when it is determined that the performance of the machine learning model trained by the second node does not meet the preset conditions; use the first learning mode to retrain the machine learning model so that the performance of the machine learning model meets the preset conditions.

[0043] In addition, according to at least one embodiment of the present invention, the method further includes:

[0044] Notifying the third node of the first learning mode;

[0045] Among them, the first learning mode is used for the third node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0046] At least one embodiment of the present invention provides an information processing device, including:

[0047] A receiving unit, configured to receive first information respectively sent by at least two second nodes; the first information represents relevant parameters of a machine learning model trained by the second node;

[0048] A first processing unit is configured to determine, for each second node, whether the performance of the machine learning model trained by the corresponding second node meets a preset condition based on the first information; and when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset condition, determine a first learning mode corresponding to the corresponding second node, and notify the corresponding second node of the first learning mode;

[0049] Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0050] At least one embodiment of the present invention provides an information processing device, including:

[0051] A second processing unit is used to infer the predicted value output by the trained machine learning model to obtain an inference result; and evaluate the inference result to obtain an evaluation result;

[0052] A third processing unit is used to determine whether to report the evaluation result to the first node; when it is determined to report the evaluation result to the first node, generate first information; the first information represents the relevant parameters of the machine learning model trained by the second node;

[0053] A sending unit is used to send the first information to the first node.

[0054] At least one embodiment of the present invention provides a first node device, including:

[0055] A first communication interface, used to receive first information respectively sent by at least two second nodes; the first information represents relevant parameters of the machine learning model trained by the second node;

[0056] A first processor is configured to determine, for each second node, based on the first information, whether the performance of the machine learning model trained by the corresponding second node meets a preset condition; and when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset condition, determine a first learning mode corresponding to the corresponding second node, and notify the corresponding second node of the first learning mode;

[0057] Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0058] At least one embodiment of the present invention provides a second node device, including:

[0059] a second processor, configured to infer the predicted value output by the trained machine learning model to obtain an inference result; evaluate the inference result to obtain an evaluation result; and determine whether to report the evaluation result to the first node; when it is determined to report the evaluation result to the first node, generate first information; the first information represents relevant parameters of the machine learning model trained by the second node;

[0060] The second communication interface is used to send the first information to the first node.

[0061] At least one embodiment of the present invention provides a first node device, characterized in that it includes a processor and a memory for storing a computer program that can be run on the processor.

[0062] Wherein, when the processor is used to run the computer program, it executes the steps of any one of the methods described above on the first node device side.

[0063] At least one embodiment of the present invention provides a second node device, characterized in that it includes a processor and a memory for storing a computer program that can be run on the processor.

[0064] Wherein, when the processor is used to run the computer program, it executes the steps of any one of the methods described above on the second node device side.

[0065] At least one embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of any of the above methods when executed by a processor.

[0066] The information processing method, device, equipment and storage medium provided by the embodiment of the present invention receive first information respectively sent by at least two second nodes; the first information represents the relevant parameters of the machine learning model trained by the second node; for each second node, based on the first information, determine whether the performance of the machine learning model trained by the corresponding second node meets the preset conditions; when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset conditions, determine the first learning mode corresponding to the corresponding second node, and notify the corresponding second node of the first learning mode; wherein the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets the preset conditions. Using the technical solution of the embodiment of the present invention, the first node can use the first information reported by the second node to determine whether the second node needs to update the machine learning model, and when it is determined that the second node needs to update the machine learning model, send the corresponding first learning mode to the second node, so that the second node can use the first learning mode to update the machine learning model, so as to use the updated machine learning model to predict the performance indicators of the network data, thereby improving the accuracy of the predicted performance indicators of the network data. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 The following is a schematic diagram of the implementation process of the information processing method of the embodiment of the present invention. Figure 1 ;

[0068] Figure 2 The following is a schematic diagram of the implementation process of the interaction between the first node and the second node in the embodiment of the present invention. Figure 1 ;

[0069] Figure 3 The following is a schematic diagram of the implementation process of the interaction between the first node and the second node in the embodiment of the present invention. Figure 2 ;

[0070] Figure 4 The following is a schematic diagram of the implementation process of the interaction between the first node and the second node in the embodiment of the present invention. Figure 3 ;

[0071] Figure 5 is a schematic diagram of physical entities corresponding to the first node and the second node in an embodiment of the present invention;

[0072] Figure 6 The following is a schematic diagram of the implementation process of the information processing method of the embodiment of the present invention. Figure 2 ;

[0073] Figure 7a and Figure 7b is a schematic diagram of modules corresponding to the first node, the second node, and the third node in an embodiment of the present invention;

[0074] Figure 8 The following is a schematic diagram of the implementation process of the interaction between the first node and the second node in the embodiment of the present invention. Figure 4 ;

[0075] Fig. 9 The structure diagram of the information processing device according to the embodiment of the present invention is as follows: Figure 1 ;

[0076] Fig.10 The structure diagram of the information processing device according to the embodiment of the present invention is as follows: Figure 2 ;

[0077] Fig.11 It is a schematic diagram of the structure of the information processing system according to an embodiment of the present invention;

[0078] Fig.12 This is a schematic diagram of the composition structure of the first node device in an embodiment of the present invention. Figure 1 ;

[0079] Fig.13 This is a schematic diagram of the composition structure of the second node device in an embodiment of the present invention. Figure 2 . DETAILED DESCRIPTION

[0080] Before introducing the technical solution of the embodiment of the present invention, the related technology is first described.

[0081] In the related art, at present, there are a large number of homogeneous network elements in mobile communication networks. The homogeneous network elements may refer to network elements with the same functions, load sharing, and strong correlation of network performance indicators. For example, network elements in the same resource pool (Pool) in the core network are homogeneous network elements, that is, network elements with the same set of performance indicators, equal number of users and load sharing, and very similar performance measurement data in most cases. Here, homogeneous network elements may specifically be a group of base stations in a wireless network that perform similarly in wireless performance data due to similar deployment scenarios.

[0082] Usually, for homogeneous network elements, in order to reduce the training workload, the same set of artificial intelligence (AI) models are often used to predict the performance indicators of network data generated in daily network operation and maintenance work, and report them to the server of the network management platform. During special maintenance work on individual network elements, such as cutover, restart, etc., the AI ​​model trained for the network element may not be applicable, and the performance of network elements in the same pool may also be affected. In addition, due to the limited storage capacity of the server, each network element can only use data stored for 1 month to train the AI ​​model. For example, based on the data collection cycle of 15 minutes, each network performance indicator can only extract 96×30=2880 sample points at a time, which is very limited for AI model training, resulting in the accuracy of the AI ​​model being affected.

[0083] In summary, in the routine maintenance of network element equipment, such as cutting and restarting, the machine learning model trained by the network element equipment may not be applicable, resulting in a decrease in the accuracy of the performance indicators of the network data predicted by the network element equipment. In addition, the storage capacity of the network management server is limited, and only one month of data can be stored at a time, which affects the training effect of the AI ​​model; if the same set of AI models is used between homogeneous network elements, it is impossible to cope with the performance changes during the operation and maintenance of individual network elements; if the models are trained separately, the training workload is huge. In the actual AI application process, data collection consumes a lot of processing resources and transmission resources. Therefore, if the frequency of data collection can be reduced or even an AI model with decent performance can be trained without collecting data, it is of practical value. At the same time, the maximum storage space of each node is also limited. When there are multiple AI training tasks at the same time, the storage space will inevitably be tight. If the consumption of storage space can be reduced as much as possible, the number of models that can be trained simultaneously by the node will increase, and the node's intelligent generation capability will be improved.

[0084] Based on this, in each embodiment of the present invention, first information respectively sent by at least two second nodes is received; the first information represents relevant parameters of the machine learning model trained by the second node; for each second node, based on the first information, it is determined whether the performance of the machine learning model trained by the corresponding second node meets the preset conditions; when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset conditions, a first learning mode corresponding to the corresponding second node is determined, and the first learning mode is notified to the corresponding second node; wherein, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets the preset conditions.

[0085] It should be noted that in the embodiment of the present invention, the AI ​​model training architecture based on transfer learning has the following advantages:

[0086] On the premise of ensuring the effect of the model, when factors such as operation and maintenance operations on a certain network element device cause changes in the performance indicators of network data using the trained machine learning model of the network element device, the first node can timely update the machine learning model of the second node, i.e., the network element device, online through the conversion of the first learning mode, thereby ensuring that the performance indicators of the network data are predicted using the updated machine learning model, thereby reducing the impact on the detection of abnormal performance indicators.

[0087] The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

[0088] The embodiment of the present invention provides an information processing method, which is applied to a first node, such as Figure 1 As shown, the method includes:

[0089] Step 101: receiving first information respectively sent by at least two second nodes; the first information represents relevant parameters of the machine learning model trained by the second node;

[0090] Step 102: For each second node, based on the first information, determine whether the performance of the machine learning model trained by the corresponding second node meets the preset conditions; when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset conditions, determine the first learning mode corresponding to the corresponding second node.

[0091] Step 103: Notify the corresponding second node of the first learning mode;

[0092] Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0093] Here, in step 101, in actual application, the first information may be an evaluation result of the second node using the trained machine learning model to evaluate and predict the index performance of the network data, and specifically may be an evaluation result obtained by the second node inferring the predicted value output by the trained machine learning model to obtain an inference result, and evaluating the inference result. After the second node reports the first information to the first node, the first node may use the first information to determine whether the machine learning model currently trained by the second node needs to be updated, so that the second node can update the currently trained machine learning model and use the updated machine learning model to predict the index performance of the network data, thereby ensuring the accuracy of the prediction result.

[0094] Here, in step 102, in actual application, the at least two second nodes may be homogeneous network elements. When the first node determines that a second node needs to update a trained machine learning model using the first information, a first learning mode corresponding to the second node is determined and notified to the second node, and the second node may retrain the machine learning model using the first learning mode to make the machine learning model predict the index performance of the network data more accurately.

[0095] The following describes in detail the process of how the first node instructs the corresponding second node to update the trained machine learning model.

[0096] In the first case, when the first node determines that a second node needs to update a trained machine learning model, the first information reported by other nodes among the at least two nodes except the second node is used to determine the first learning mode used by the second node to retrain the machine learning model.

[0097] In actual application, after the at least two second nodes report their respective first information to the first node, the first node can use the first information to determine whether the performance of the machine learning model trained by the second node meets the preset conditions, that is, whether the result of predicting the indicator performance of the network data using the trained machine learning model is within the preset range. When it is determined that the performance of the machine learning model trained by a second node does not meet the preset conditions, the learning model used by other nodes to train the machine learning model can be used to update the learning model used by the second node to train the machine learning model.

[0098] Based on this, in one embodiment, the first information includes a second learning mode of the second node training the machine learning model; and determining the first learning mode corresponding to the corresponding second node includes:

[0099] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode;

[0100] When it is determined that the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain a plurality of nodes to be processed;

[0101] Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result;

[0102] Determine the machine learning model with the best performance evaluation in the evaluation results; use the transfer learning mode corresponding to the machine learning model with the best performance evaluation as the first learning mode corresponding to the corresponding second node;

[0103] The transfer learning mode characterizes that the parameters of the machine learning model are obtained through transfer learning.

[0104] Here, the first information may also include an indicator performance prediction result of the second node predicting the indicator performance of the network data using a machine learning model. In this way, the first node may compare the indicator performance prediction results corresponding to the multiple nodes to be processed with the preset indicator threshold to obtain a comparison result, and select from the comparison result the node whose performance indicator prediction result is closest to the preset indicator threshold as the node corresponding to the machine learning model with the best performance evaluation.

[0105] For example, Table 1 is a schematic diagram of the first information reported by the second node to the first node. As shown in Table 1, assuming that the first node uses the first information reported by the second node 5 to determine that the performance of the machine learning model trained by the node does not meet the preset conditions, the first node can use the first information to determine that the learning mode of the second node 1, the second node 2 and the second node 3 is the transfer learning mode, and use the second node 1, the second node 2 and the second node 3 as the nodes to be processed. Using the indicator performance prediction results in the first information and combined with the preset indicator threshold, the performance of the machine learning models trained by the second node 1, the second node 2 and the second node 3 can be evaluated and ranked, and the node corresponding to the machine learning model with the best performance evaluation is obtained, assuming it is the second node 1, as shown in Table 2, then the transfer learning mode corresponding to the second node 1 is used as the first learning mode corresponding to the second node 5 and notified to the second node 5.

[0106] Second node number First Information Second Node 1 Transfer learning model, indicator performance prediction results 1 Second Node 2 Transfer learning model, indicator performance prediction results 2 Second Node 3 Transfer learning model, indicator performance prediction results 3 Second Node 4 Autonomous learning model, indicator performance prediction results 4 Second Node 5 Autonomous learning model, indicator performance prediction results 5

[0107] Table 1

[0108] Second node number Performance evaluation ranking Second Node 1 Best performance evaluation Second Node 2 Performance Evaluation II Second Node 3 Performance Evaluation Third

[0109] Table 2

[0110] In actual application, after at least two second nodes report their respective first information to the first node, the first information can be used to determine the second node that needs to retrain the machine learning model, and the first information reported by other nodes among the at least two second nodes except the second node can be used to determine the learning mode used by the second node to retrain the machine learning model.

[0111] Based on this, in one embodiment, notifying the corresponding second node of the first learning mode includes:

[0112] Generate second information; the second information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model;

[0113] The second information is sent to the corresponding second node.

[0114] In an example, taking the first node as a CU and the second node as a DU as an example, the interaction process between the first node and the second node is described, such as Figure 2 As shown, including:

[0115] Step 201: CU receives first information sent by DU-A and DU-B respectively.

[0116] Here, CU serves as the first node, which is used to aggregate and evaluate the first information reported by all distributed DU nodes, namely, the parameters of the machine learning model trained by the DU node and the indicator performance evaluation results of predicting the indicator performance of network data using the corresponding machine learning model; and generate new learning mode recommendations for each DU node.

[0117] Step 202: For DU-A, the CU determines whether the performance of the machine learning model trained by DU-A meets the preset conditions; when it is determined that the performance of the machine learning model trained by DU-A does not meet the preset conditions, execute step 203.

[0118] Here, when CU determines that the performance of the machine learning model trained by DU-A does not meet the preset conditions, it can determine whether the learning mode of the machine learning model trained by DU-B is the transfer learning mode. Assuming that the learning mode of the machine learning model trained by DU-B is the transfer learning mode, the learning mode used by the machine learning model trained by DU-B is recommended to DU-A for use.

[0119] Step 203: Determine that the first learning mode corresponding to DU-A is the transfer learning mode; generate second information; and send the second information to DU-A.

[0120] Here, the second information may specifically be: a Learning mode instruction message, the content of which may include the name or number of the DU-A node, the recommended learning mode being the transfer learning mode, the source machine learning model to be migrated, the name / number of the node corresponding to the source machine learning model, the number of the source machine learning model, etc. Among them, the source machine learning model to be migrated is the machine learning model trained by DU-B, for example, 3×3, W, D, wherein 3×3 indicates that the machine learning model trained by DU-B includes 3 layers, W indicates the weight of the model, and D indicates the weight of the model.

[0121] Here, after receiving the Learning mode instruction message, the DU-A node can update the learning mode according to the instructions and retrain the machine learning model according to the new learning mode; it can also save the migrated source machine learning model and its corresponding node name / number, the number of the source machine learning model, etc.

[0122] Here, the process for DU-B is similar to that for DU-A and will not be described in detail here.

[0123] Here, CU, as a functional entity, can implement the following functions:

[0124] 1. According to the compliance conditions, evaluate the performance of the machine learning model reported by each DU node. If the machine learning model trained by a DU node does not meet the standards, generate a new learning mode recommendation for the DU node and send it to the corresponding DU node through the learning mode instruction message.

[0125] 2. The content of the Learning mode instruction message may include the name or number of the node, the learning mode recommended for the node (including transfer learning), the source machine learning model to be migrated (including a machine learning model), the node name / number corresponding to the source machine learning model, and the number of the source machine learning model.

[0126] 3. According to the compliance conditions, evaluate the performance of the machine learning model reported by each DU node. If the machine learning model trained by a certain DU node meets the standards, and there are multiple such models, select the recommended model for the DU node according to the strategy from all the qualified machine learning models, and send it to the corresponding DU node through the Online model instruction message. After receiving the Online model instruction message, the online reasoning module of the DU node uses the model indicated in the message for online reasoning. If the message includes the generation time of the machine learning model, the generation time of the machine learning model can be used to determine whether the model is invalid. If it is not invalid, it will be used; if it is invalid, the model will be selected as the online reasoning model. Among them, the online reasoning module of the DU node can also delete other unused models.

[0127] 4. The content of the Online model instruction message can include the node name, the number of the machine learning model, and the generation time of the machine learning model.

[0128] 5. You can configure the model performance compliance judgment conditions, for example, judging whether the machine learning model trained by the DU node meets the standards based on the preset performance threshold.

[0129] 6. You can configure the model selection strategy. For example, you can select the machine learning model with the best performance as the source machine learning model for migration, or you can select the machine learning model trained on the node with the least data storage as the source machine learning model for migration, and so on.

[0130] In the second case, when the first node determines that a second node needs to update the trained machine learning model, the first learning mode used to train the machine learning model of the second node can be used for re-training.

[0131] In actual application, after the at least two second nodes report their respective first information to the first node, the first node can use the first information to determine whether the performance of the machine learning model trained by the second node meets the preset conditions, that is, whether the result of predicting the indicator performance of the network data using the trained machine learning model is within the preset threshold range. When it is determined that the performance of the machine learning model trained by a second node does not meet the preset conditions, the learning model used to train the machine learning model of the second node can be used for re-training.

[0132] Based on this, in one embodiment, the first information includes a second learning mode of the second node training the machine learning model; and determining the first learning mode corresponding to the corresponding second node includes:

[0133] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning modes of the machine learning models trained by the other nodes are not transfer learning modes;

[0134] When it is determined that the learning modes of the machine learning models trained by the other nodes are not transfer learning modes, the autonomous learning mode is used as the first learning mode corresponding to the corresponding second node;

[0135] Among them, the autonomous learning mode represents that the corresponding second node keeps the model parameters unchanged when retraining the machine learning model.

[0136] For example, Table 3 is a schematic diagram of the first information reported by the second node to the first node. As shown in Table 3, assuming that the first node uses the first information reported by the second node 5 to determine that the performance of the machine learning model trained by the node does not meet the preset conditions, and determines that the learning modes of the second node 1, the second node 2, the second node 3 and the second node 4 are not transfer learning modes, then the autonomous learning mode is notified to the second node 5.

[0137] Second node number First Information Second Node 1 Autonomous learning model, indicator performance prediction results 1 Second Node 2 Autonomous learning model, indicator performance prediction results 2 Second Node 3 Autonomous learning model, indicator performance prediction results 3 Second Node 4 Autonomous learning model, indicator performance prediction results 4 Second Node 5 Autonomous learning model, indicator performance prediction results 5

[0138] Table 3

[0139] In actual application, after at least two second nodes report their respective first information to the first node, the first node can use the first information to determine the second node that needs to retrain the machine learning model, and determine that the learning mode used by the second node to retrain the machine learning model is the autonomous learning mode.

[0140] Based on this, in one embodiment, notifying the corresponding second node of the first learning mode includes:

[0141] generating third information; the third information including the name or number of the corresponding second node and the recommended first learning mode;

[0142] The third information is sent to the corresponding second node.

[0143] Here, the recommended first learning mode is the autonomous learning mode, that is, the corresponding second node keeps the model parameters unchanged when retraining the machine learning model, for example, using the re-selected network data to retrain the machine learning model.

[0144] In an example, taking the first node as a CU and the second node as a DU as an example, the interaction process between the first node and the second node is described, such as Figure 3 As shown, including:

[0145] Step 301: CU receives first information sent by DU-A and DU-B respectively.

[0146] Here, CU serves as the first node, which is used to aggregate and evaluate the first information reported by all distributed DU nodes, namely, the parameters of the machine learning model trained by the DU node and the indicator performance evaluation results of predicting the indicator performance of network data using the corresponding machine learning model; and generate new learning mode recommendations for each DU node.

[0147] Step 302: For DU-A, the CU determines whether the performance of the machine learning model trained by DU-A meets the preset conditions; when it is determined that the performance of the machine learning model trained by DU-A does not meet the preset conditions, execute step 303.

[0148] Here, when CU determines that the performance of the machine learning model trained by DU-A does not meet the preset conditions, it can determine whether the learning mode of the machine learning model trained by DU-B is the transfer learning mode. Assuming that the learning mode of the machine learning model trained by DU-B is not the transfer learning mode, the learning mode used by DU-A to retrain the machine learning model is set to the autonomous learning mode.

[0149] Step 303: Determine that the first learning mode corresponding to DU-A is the autonomous learning mode; generate third information; and send the third information to DU-A.

[0150] Here, the third information may specifically be: a Learning mode instruction message, and the content of the Learning mode instruction message may include the name or number of the DU-A node, and the recommended learning mode is the autonomous learning mode.

[0151] Here, after receiving the Learning mode instruction message, the DU-A node can update the learning mode according to the instructions and retrain the machine learning model according to the autonomous learning mode.

[0152] Here, the process for DU-B is similar to that for DU-A and will not be described in detail here.

[0153] Here, the functional entity corresponding to the first node interacts with the functional entity corresponding to the second node, which has the following advantages:

[0154] (1) While ensuring the model effect, reduce the amount of data reporting or storage space requirements; or improve model performance under the premise of a certain total amount of data storage space.

[0155] (2) By switching the learning mode, it is ensured that when the indicator model changes due to operation and maintenance operations on a certain network element, the online models of all network elements can be updated in time to reduce the impact on model performance.

[0156] In the third case, when the first node determines that a second node needs to update the trained machine learning model, the first information reported by other nodes among the at least two nodes except the second node is used to determine the first learning mode used by the second node to retrain the machine learning model.

[0157] In actual application, after the at least two second nodes report their respective first information to the first node, the first node can use the first information to determine whether the performance of the machine learning model trained by the second node meets the preset conditions, that is, whether the result of predicting the indicator performance of the network data using the trained machine learning model is within the preset range. When it is determined that the performance of the machine learning model trained by a second node does not meet the preset conditions, the learning model used by other nodes to train the machine learning model and the learning model used by the second stage to train the machine learning model can be used to update the learning model used to train the machine learning model of the second node.

[0158] Based on this, in one embodiment, the first information includes a second learning mode of the second node training the machine learning model; and determining the first learning mode corresponding to the corresponding second node includes:

[0159] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the learning mode of the machine learning model trained by the other nodes is a transfer learning mode;

[0160] When it is determined that the learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain multiple nodes to be processed;

[0161] Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result;

[0162] The combination of the transfer learning mode and the autonomous learning mode corresponding to the machine learning model whose performance evaluation in the evaluation result is greater than the performance threshold is used as the first learning mode corresponding to the corresponding second node;

[0163] Among them, the combination of the transfer learning mode and the autonomous learning mode represents that the corresponding second node retrains the machine learning model while keeping the model parameters unchanged and while updating the model parameters.

[0164] For example, Table 4 is a schematic diagram of the first information reported by the second node to the first node. As shown in Table 3, it is assumed that the first node uses the first information reported by the second node 5 to determine that the performance of the machine learning model trained by the node does not meet the preset conditions, and determines that the learning modes of the second node 1 and the second node 4 are not transfer learning modes, and the learning modes of the second node 2 and the second node 3 are migration modes. If the indicator performance prediction results corresponding to the second node 2 and the second node 3 are respectively greater than the performance threshold, the migration learning mode used by the second node 2 to train the machine learning model and the migration learning mode used by the second node 3 to train the machine learning model are combined with the autonomous learning mode, and the combined learning mode is notified to the second node 5 so that the second node 5 can retrain its own machine learning model according to the recommended learning mode.

[0165] Second node number First Information Second Node 1 Autonomous learning model, indicator performance prediction results 1 Second Node 2 Transfer learning model, indicator performance prediction results 2 Second Node 3 Transfer learning model, indicator performance prediction results 3 Second Node 4 Autonomous learning model, indicator performance prediction results 4 Second Node 5 Autonomous learning model, indicator performance prediction results 5

[0166] Table 4

[0167] In actual application, after at least two second nodes report their respective first information to the first node, the first node can use the first information to determine the second node that needs to retrain the machine learning model, and determine that the learning mode used by the second node to retrain the machine learning model is a combination of the autonomous learning mode and the transfer learning mode.

[0168] In actual application, the sending to the corresponding second node includes:

[0169] Generate fourth information; the fourth information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model;

[0170] The fourth information is sent to the corresponding second node.

[0171] Here, the recommended first learning mode is a combination of an autonomous learning mode and a transfer learning mode, that is, the corresponding second node retrains the machine learning model while keeping the model parameters unchanged, and retrains the machine learning model after updating the model parameters of other indicated second nodes.

[0172] In an example, taking the first node as a CU and the second node as a DU as an example, the interaction process between the first node and the second node is described, such as Figure 4 As shown, including:

[0173] Step 401: CU receives first information sent by DU-A, DU-B and DU-C respectively.

[0174] Here, CU serves as the first node, which is used to aggregate and evaluate the first information reported by all distributed DU nodes, namely, the parameters of the machine learning model trained by the DU node and the indicator performance evaluation results of predicting the indicator performance of network data using the corresponding machine learning model; and generate new learning mode recommendations for each DU node.

[0175] Step 402: For DU-A, the CU determines whether the performance of the machine learning model trained by DU-A meets the preset conditions; when it is determined that the performance of the machine learning model trained by DU-A does not meet the preset conditions, execute step 403.

[0176] Here, when CU determines that the performance of the machine learning model trained by DU-A does not meet the preset conditions, it can determine whether the learning modes of the machine learning models trained by DU-B and DU-C are both transfer learning modes. Assuming that the learning modes of the machine learning models trained by DU-B and DU-C are both transfer learning modes, and the indicator performance prediction results reported by DU-B and DU-C respectively are greater than the performance threshold, then the transfer learning mode of DU-B, the transfer learning mode of DU-C and the autonomous learning mode are combined, and the learning mode used by DU-A to retrain the machine learning model is set to a combination of the transfer learning mode of DU-B, the transfer learning mode of DU-C and the autonomous learning mode.

[0177] Step 403: Determine that the first learning mode corresponding to DU-A is a combination of the autonomous learning mode and the transfer learning mode; generate fourth information, and send the fourth information to DU-A.

[0178] Here, the fourth information may specifically be: a Learning mode instruction message, and the content of the Learning mode instruction message may include the name or number of the DU-A node, and the recommended learning mode is a combination of an autonomous learning mode and a transfer learning mode.

[0179] Here, after receiving the Learning mode instruction message, the DU-A node can update the learning mode according to the instructions and retrain the machine learning model according to the autonomous learning mode.

[0180] Here, the process for DU-B and DU-C is similar to that for DU-A, and will not be repeated here.

[0181] Here, the functional entity corresponding to the first node interacts with the functional entity corresponding to the second node, which has the following advantages:

[0182] (1) While ensuring the model effect, reduce the amount of data reporting or storage space requirements; or improve model performance under the premise of a certain total amount of data storage space.

[0183] (2) By switching the learning mode, it is ensured that when the indicator model changes due to operation and maintenance operations on a certain network element, the online models of all network elements can be updated in time to reduce the impact on model performance.

[0184] The implementation process of the information processing method according to the embodiment of the present invention is described in detail below in conjunction with specific embodiments.

[0185] Figure 5 is a schematic diagram of physical entities corresponding to the first node and the second node in an embodiment of the present invention, such as Figure 5 As shown, the system includes:

[0186] The two base stations, denoted by DU A and DU B respectively, are used to send relevant parameters of the machine learning model trained by themselves, such as performance indicator evaluation results, to the CU.

[0187] A central equipment base station, represented by CU, may include a "model evaluation" entity, which is used to evaluate whether the performance of the machine learning model trained by each DU meets the requirements.

[0188] Specifically, for DU B, if it has a qualified model, a recommended model is selected for the node from among all qualified models according to the strategy, and the model is indicated to the B base station through a message;

[0189] For DU A, its initial learning mode is autonomous learning. If it is tracked that it has a stable model that meets the standard for a long time, its learning mode will be changed to autonomous + transfer learning (the source model is DU B's model). At this time, DU A will generate a new transfer learning model and send the evaluation results to the model evaluation entity corresponding to CU. If the performance of the transfer learning model meets the standard, DU A can be instructed to adopt transfer learning.

[0190] Due to the dynamic changes in the scene, the performance of the transfer learning model used by DU A has declined and no longer meets the standard. The model evaluation entity corresponding to CU instructs to change its learning mode to autonomous + transfer learning; at this time, DU A will generate a new autonomous learning model and send the evaluation results to the model evaluation entity. If the performance of the autonomous learning model meets the standard, DU A can be instructed to adopt autonomous learning.

[0191] By adopting the technical solution provided by the embodiment of the present invention, the first node can use the first information reported by the second node to determine whether the second node needs to update the machine learning model, and when it is determined that the second node needs to update the machine learning model, the corresponding first learning model is sent to the second node. In this way, the second node can use the first learning model to update the machine learning model, and use the updated machine learning model to predict the performance indicators of the network data, thereby ensuring the accuracy of the predicted performance indicators of the network data.

[0192] The embodiment of the present invention also provides an information processing method, which is applied to a second node, such as Figure 6 As shown, the method includes:

[0193] Step 601: inferring the predicted value output by the trained machine learning model to obtain an inference result; evaluating the inference result to obtain an evaluation result;

[0194] Step 602: Determine whether to report the evaluation result to the first node; when it is determined to report the evaluation result to the first node, generate first information; the first information represents relevant parameters of the machine learning model trained by the second node;

[0195] Step 603: Send the first information to the first node.

[0196] Here, in step 601, the second node infers the predicted value output by the trained machine learning model to obtain an inference result, and evaluates the inference result to obtain an evaluation result. Subsequently, first information can be generated based on the evaluation result and reported to the first node.

[0197] Here, in step 602, after the second node reports the first information to the first node, the first node can use the first information to determine whether the machine learning model currently trained by the second node needs to be updated, so that the second node can update the currently trained machine learning model and use the updated machine learning model to predict the indicator performance of the network data to ensure the accuracy of the prediction results.

[0198] In actual application, the second node can report the learning mode of the trained machine learning model to the first node. In this way, when the first node determines that the learning mode of the machine learning model of a second node needs to be updated, it can determine the model selection strategy according to the learning mode reported by the second node.

[0199] Based on this, in one embodiment, generating the first information includes:

[0200] The first information is generated based on the name or number of the second node, and the learning mode and parameters of the machine learning model trained by the second node.

[0201] In actual application, when the first node determines to update the learning mode of the machine learning model of the second node, the second node can receive the first learning mode sent by the first node and retrain the machine learning model using the first learning mode.

[0202] Based on this, in one embodiment, the method further includes:

[0203] Receiving a first learning mode sent by the first node; the first learning mode is determined by the first node when it is determined that the performance of the machine learning model trained by the second node does not meet a preset condition;

[0204] The machine learning model is retrained using the first learning mode so that the performance of the machine learning model meets preset conditions.

[0205] Here, the first learning mode includes at least one of the following:

[0206] Transfer learning model,

[0207] Self-learning mode,

[0208] A combination of transfer learning mode and autonomous learning mode.

[0209] Here, the second node may receive the first learning mode sent by the first node through a Learning mode instruction message, which specifically includes:

[0210] When the first learning mode is the transfer learning mode, the content of the Learning mode instruction message may include the name or number of the DU-A node, the recommended learning mode is the transfer learning mode, the source machine learning model to be migrated, the name / number of the node corresponding to the source machine learning model, the number of the source machine learning model, etc.

[0211] When the first learning mode is the autonomous learning mode, the content of the Learning mode instruction message may include the name or number of the DU-A node and the recommended learning mode is the autonomous learning mode.

[0212] When the first learning mode is a combination of the transfer learning mode and the autonomous learning mode, the content of the Learning mode instruction message may include the name or number of the DU-A node and the recommended learning mode is a combination of the autonomous learning mode and the transfer learning mode.

[0213] In actual application, in addition to updating the learning mode of the machine learning model by itself, the second node can also update the learning mode of the machine learning model through the third node.

[0214] Based on this, in one embodiment, the method further includes:

[0215] Notifying the third node of the first learning mode;

[0216] Among them, the first learning mode is used for the third node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0217] Here, in actual application, the third node and the second node may be two separate physical entities, or may be a centralized physical entity.

[0218] Here, the first node may also notify the third node of the first learning mode.

[0219] Figure 7a and Figure 7b is a schematic diagram of the modules corresponding to the first node, the second node, and the third node, such as Figure 7a As shown, the first node is provided with a model evaluation module, the second node is provided with an online reasoning module, and the third node is provided with an AI training module.

[0220] The online reasoning module of the second node is used to perform online reasoning on all currently selected models based on real-time samples, evaluate the performance of the model according to the reasoning results, and upload the model and evaluation results to the first node.

[0221] The model evaluation module of the first node is used to determine which nodes have qualified models available and which nodes do not have qualified models based on the model performance evaluation results and set thresholds. For qualified nodes, the AI ​​training module is instructed to recommend the model and learning mode (multiple) to be used, and the similarity of the models between nodes is analyzed. For nodes with similar models, some nodes are equipped with migration modes at the same time; for nodes that do not meet the standards, a new learning mode (multiple) for the node is recommended to the AI ​​training module.

[0222] The AI ​​training module of the third node is used to generate a set of usable models for each second node according to the learning mode recommendation instructions of the first node, wherein the model can be only an autonomous learning model, or multiple transfer learning models, or an autonomous model + a transfer learning model. The models can also be stored in chronological order, and according to the instructions of the first node, the currently suitable model can be selected for use.

[0223] like Figure 7bAs shown, the AI ​​training entity and online reasoning entity of each DU are scattered in different physical entities, the AI ​​training module is a centralized AI server, and the model evaluation module is located in the CU.

[0224] In one example, taking the first node as CU, the second node as DU, and the third node as AI training node as an example, the interaction process between the first node and the second node is described, such as Figure 8 As shown, including:

[0225] Step 801: DU-A infers the predicted value output by the trained machine learning model to obtain an inference result; and evaluates the inference result to obtain an evaluation result.

[0226] Here, DU-A is the second node.

[0227] Step 802: Determine whether to report the evaluation result to the CU; when it is determined to report the evaluation result to the CU, execute step 803.

[0228] Step 803: Generate first information; and send it to CU.

[0229] Here, CU is the first node. DU-A can configure the trigger conditions for uploading the evaluation results, including periodicity and event, for example, the model performance change exceeds the threshold, or the absolute value of the model performance exceeds the threshold; when the trigger conditions for uploading the evaluation results are met, the first information is reported to CU.

[0230] Here, the first information may specifically be a Model evaluation results update message. The content of the Model evaluation results update may include: node name, model number, model content, model learning mode (autonomous or migration), model evaluation results, and model generation time; if it is a migration model, it also includes the node name / number of the migrated source machine learning model and the number of the migrated source machine learning model.

[0231] Here, after receiving the first information, the CU can save the performance evaluation result of the corresponding DU-A contained in the first information and update the current set of all model evaluation results; if the message content includes the model generation time, the old model can be deleted according to the time.

[0232] Step 804: Receive a first learning mode sent by the CU; and notify the AI ​​training node of the first learning mode;

[0233] Among them, the first learning mode is used for the AI ​​training node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0234] Here, DU-A, as a functional entity, has the following functions:

[0235] 1. You can configure the current valid model time interval or number or model learning mode or model expiration time, filter machine learning models according to the configuration conditions, and perform inference on all machine learning models that meet the conditions, that is, infer the predicted values ​​output by the trained machine learning models to obtain inference results.

[0236] 2. You can configure the evaluation cycle and evaluate the inference results according to the configured evaluation cycle. For example, you can evaluate the machine learning model with the best inference results.

[0237] 3. According to the first learning mode sent by CU, configure the online reasoning model, synchronize the configured online reasoning model information to its own AI training module through the Model update response message, and retrain the machine learning model by its own AI training module. After receiving the Model update response message, the AI ​​training module of DU-A can only retain the machine learning model used by the node configuration and delete other machine learning models. The content of the Model update response message may include the node name, model number, and model generation time.

[0238] Here, the AI ​​training node is a functional entity with the following functions:

[0239] (1) The model is generated according to the self-learning mode by default

[0240] (2) Receive and parse the Learning mode instruction message, and generate one or more models for the node according to the instructed learning mode, including autonomous learning only (generating one model), transfer learning only (generating one or more transfer models), and autonomous + transfer learning (generating multiple models).

[0241] (3) Synchronize the latest model to the corresponding online reasoning module through the following message:

[0242] Model Update request: The content includes node name, task number, model number, model content, model learning mode (autonomous or migration), and if it is a migration model, it also includes the migration model source node name / number, migration source model number, and can also include the model generation time.

[0243] (4) After receiving the message, the online reasoning module saves all the above model-related information under the node name.

[0244] The process of interaction between the first node and the second node is described in detail below with reference to specific embodiments.

[0245] In the following content, the AI ​​training server corresponds to the third node, DU-A and DU-B correspond to the second node, and CU corresponds to the first node.

[0246] The AI ​​training server stores data samples of network performance indicators KPI (such as switching success rate). The AI ​​training server trains a prediction model that can predict the switching success rate based on the stored samples. The stored model set is shown in Table 5.

[0247]

[0248] Table 5

[0249] The AI ​​training server updates the above model to DU-A and DU-B. The relevant message format is as follows:

[0250] Model update request message: {node name: DU_A; task number: predict switching success rate; model 1: {model number: m_A_s_1; learning mode: autonomous learning; generation time: 2020-10-22hh:mm:ss; model content: ...}; model 2: { ...}}

[0251] DU_B uses the m_B_s_1 model for online reasoning and calculates the performance of the model, for example, the prediction accuracy is 92%, and sends the result to CU. The message content example is as follows:

[0252] Model evaluation result update: {node name: DU_B; task number: predict switching success rate;

[0253] Model 1: {Model number: m_B_s_1; Learning mode: autonomous learning; Generation time: 2020-10-22hh:mm:ss; Model content: ..., Evaluation result: 92%}

[0254] }

[0255] DU_A uses the m_A_s_1 and m_A_B_s_1 models for online reasoning, and calculates the performance of the models, for example, the prediction accuracy is 93% and 91% respectively, and sends the result to CU. The message content example is as follows:

[0256] Model evaluation result update: {node name: DU_A; task number: predict switching success rate;

[0257] Model 1: {Model number: m_A_s_1; Learning mode: autonomous learning; Generation time: 2020-10-22hh:mm:ss; Model content: ..., Evaluation result: 95%};

[0258] Model 2: {Model number: m_A_B_s_1; Learning mode: Transfer learning; Generation time: 2020-10-22hh:mm:ss; Model content: ..., Evaluation result: 91%}

[0259] }

[0260] CU evaluates the prediction accuracy threshold of 90% and believes that DU_B's model m_B_s_1 has met the standard. Both models of DU_A have met the standard. Further, based on the policy "minimum data storage", model m_A_B_s_1 is selected for DU_A and notified to DU through a message. The message content example is as follows:

[0261] Online Model Instructions:

[0262] {Node name: DU_B; Task number: Predicted switching success rate; Recommended model: {Model number: m_B_s_1; ...}}

[0263] {Node name: DU_A; Task number: Predicted switching success rate; Recommended model: {Model number: m_A_B_s_1; ...}}

[0264] If the prediction accuracy threshold configured by CU is 93%, it is considered that DU_B does not have a qualified model, and the learning mode of B is changed to "autonomous + transfer learning" through a message. The message content example is as follows:

[0265] Learning mode instructions:

[0266] {Node name: DU_B; Task number: Predict switching success rate; Learning mode: {Autonomous learning, transfer learning: {Source model: m_A_s_1, Source model performance: 95%}}}

[0267] After receiving the learning mode instruction, the AI ​​training server learns a new model m_B_A_s_1 for DU_B according to the migrated machine learning model, as follows:

[0268]

[0269] Table 6

[0270] By adopting the technical solution of the embodiment of the present invention, the first node can use the first information reported by the second node to determine whether the second node needs to update the machine learning model, and when it is determined that the second node needs to update the machine learning model, the corresponding first learning model is sent to the second node. In this way, the second node can use the first learning model to update the machine learning model, and use the updated machine learning model to predict the performance indicators of the network data, thereby ensuring the accuracy of the predicted performance indicators of the network data.

[0271] In order to implement the information processing method of the embodiment of the present invention, the embodiment of the present invention also provides an information processing device, Fig. 9 Schematic diagram of the structure of the information processing device according to the embodiment of the present invention; Fig. 9 As shown, the device comprises:

[0272] A receiving unit 91 is configured to receive first information respectively sent by at least two second nodes; the first information represents relevant parameters of a machine learning model trained by the second node;

[0273] A first processing unit 92 is configured to determine, for each second node, whether the performance of the machine learning model trained by the corresponding second node meets a preset condition based on the first information; and when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset condition, determine a first learning mode corresponding to the corresponding second node, and notify the corresponding second node of the first learning mode;

[0274] Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0275] In one embodiment, the first processing unit 92 is specifically configured to: the first information includes a second learning mode in which the second node trains a machine learning model;

[0276] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode;

[0277] When it is determined that the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain a plurality of nodes to be processed;

[0278] Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result;

[0279] Determine the machine learning model with the best performance evaluation in the evaluation results; use the transfer learning mode corresponding to the machine learning model with the best performance evaluation as the first learning mode corresponding to the corresponding second node;

[0280] The transfer learning mode characterizes that the parameters of the machine learning model are obtained through transfer learning.

[0281] In one embodiment, the first processing unit 92 is specifically configured to:

[0282] Generate second information; the second information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model;

[0283] The second information is sent to the corresponding second node.

[0284] In addition, according to at least one embodiment of the present invention, the first processing unit 92 is specifically configured to:

[0285] The first information includes a second learning mode in which the second node trains a machine learning model;

[0286] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning modes of the machine learning models trained by the other nodes are not transfer learning modes;

[0287] When it is determined that the learning modes of the machine learning models trained by the other nodes are not transfer learning modes, the autonomous learning mode is used as the first learning mode corresponding to the corresponding second node;

[0288] Among them, the autonomous learning mode represents that the corresponding second node keeps the model parameters unchanged when retraining the machine learning model.

[0289] In one embodiment, the first processing unit 92 is specifically configured to: generate third information; the third information includes the name or number of the corresponding second node and the recommended first learning mode;

[0290] The third information is sent to the corresponding second node.

[0291] In addition, according to at least one embodiment of the present invention, the first information includes a second learning mode of the second node training the machine learning model; the first processing unit 92 is specifically configured to:

[0292] For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the learning mode of the machine learning model trained by the other nodes is a transfer learning mode;

[0293] When it is determined that the learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain multiple nodes to be processed;

[0294] Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result;

[0295] The combination of the transfer learning mode and the autonomous learning mode corresponding to the machine learning model whose performance evaluation in the evaluation result is greater than the performance threshold is used as the first learning mode corresponding to the corresponding second node;

[0296] Among them, the combination of the transfer learning mode and the autonomous learning mode represents that the corresponding second node retrains the machine learning model while keeping the model parameters unchanged and while updating the model parameters.

[0297] In one embodiment, the first processing unit 92 is specifically used to: generate fourth information; the fourth information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model;

[0298] The fourth information is sent to the corresponding second node.

[0299] In actual application, the receiving unit 91 can be implemented by a communication interface in an information processing device; the first processing unit 92 can be implemented by a processor in the information processing device.

[0300] It should be noted that: when the information processing device provided in the above embodiment performs information processing, only the division of the above program modules is used as an example. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the information processing device provided in the above embodiment and the information processing method embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0301] In order to implement the information processing method of the embodiment of the present invention, the embodiment of the present invention also provides an information processing device, Fig.10 Schematic diagram of the structure of the information processing device according to the embodiment of the present invention; Fig.10 As shown, the device comprises:

[0302] The second processing unit 101 is used to infer the predicted value output by the trained machine learning model to obtain an inference result; and evaluate the inference result to obtain an evaluation result;

[0303] The third processing unit 102 is used to determine whether to report the evaluation result to the first node; when it is determined to report the evaluation result to the first node, generate first information; the first information represents the relevant parameters of the machine learning model trained by the second node;

[0304] The sending unit 103 is configured to send the first information to the first node.

[0305] In addition, according to at least one embodiment of the present invention, the third processing unit 102 is specifically configured to:

[0306] The first information is generated based on the name or number of the second node, and the learning mode and parameters of the machine learning model trained by the second node.

[0307] In one embodiment, the device further comprises:

[0308] A training unit is used to receive a first learning mode sent by the first node; the first learning mode is determined by the first node when it is determined that the performance of the machine learning model trained by the second node does not meet the preset conditions; and the machine learning model is retrained using the first learning mode to make the performance of the machine learning model meet the preset conditions.

[0309] In one embodiment, the sending unit 103 is further configured to:

[0310] Notifying the third node of the first learning mode;

[0311] Among them, the first learning mode is used for the third node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

[0312] In actual application, the second processing unit 101 and the third processing unit 102 can be implemented by a processor in an information processing device. The sending unit 103 can be implemented by a processor in an information processing device.

[0313] It should be noted that: when the information processing device provided in the above embodiment performs information processing, only the division of the above program modules is used as an example. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the information processing device provided in the above embodiment and the information processing method embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0314] In order to implement the information processing method of the embodiment of the present invention, the embodiment of the present invention also provides an information processing system. Fig.11 FIG. 1 is a schematic diagram of the structure of the information processing system according to an embodiment of the present invention; Fig.10 As shown, the system comprises:

[0315] The second node 111 is used to infer the predicted value output by the trained machine learning model to obtain an inference result;

[0316] Evaluate the inference result to obtain an evaluation result; determine whether to report the evaluation result to the first node; when it is determined to report the evaluation result to the first node, generate first information; the first information represents relevant parameters of the machine learning model trained by the second node; and send the first information to the first node.

[0317] The first node 112 is used to receive first information sent by at least two second nodes respectively; the first information represents relevant parameters of the machine learning model trained by the second node; for each second node, based on the first information, determine whether the performance of the machine learning model trained by the corresponding second node meets the preset conditions; when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset conditions, determine a first learning mode corresponding to the corresponding second node, and notify the corresponding second node of the first learning mode; wherein, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets the preset conditions.

[0318] Here, the process of the first node 112 and the second node 111 performing information processing has been described above and will not be repeated here.

[0319] The embodiment of the present invention further provides a first node device, such as Fig.12 As shown, including:

[0320] The first communication interface 121 is capable of exchanging information with other devices;

[0321] The first processor 122 is connected to the first communication interface 121 and is used to execute the method provided by one or more technical solutions on the smart device side when running a computer program. The computer program is stored in the first memory 123.

[0322] It should be noted that the specific processing process of the first processor 122 and the first communication interface 121 is detailed in the method embodiment and will not be repeated here.

[0323] Of course, in actual application, the various components in the first node device 120 are coupled together through the bus system 124. It can be understood that the bus system 124 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 124 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Fig.12 Various buses are labeled as bus system 124 .

[0324] The first memory 123 in the embodiment of the present application is used to store various types of data to support the operation of the first node device 120. Examples of such data include: any computer program used to operate on the first node device 120.

[0325] The method disclosed in the above embodiment of the present application can be applied to the first processor 122, or implemented by the first processor 122. The first processor 122 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the first processor 122. The above-mentioned first processor 122 may be a general-purpose processor, a digital data processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The first processor 122 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present application, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in the first memory 123, and the first processor 122 reads the information in the first memory 123 and completes the steps of the above method in combination with its hardware.

[0326] The embodiment of the present invention further provides a second node device, such as Fig.13 As shown, including:

[0327] The second communication interface 131 is capable of exchanging information with other devices;

[0328] The second processor 132 is connected to the second communication interface 131 and is used to execute the method provided by one or more technical solutions on the smart device side when running a computer program. The computer program is stored in the second memory 133.

[0329] It should be noted that: the specific processing process of the second processor 132 and the second communication interface 131 is detailed in the method embodiment, which will not be repeated here.

[0330] Of course, in actual application, the various components in the second node device 130 are coupled together through the bus system 134. It can be understood that the bus system 134 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 134 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Fig.13 Various buses are labeled as bus system 134 .

[0331] The second memory 133 in the embodiment of the present application is used to store various types of data to support the operation of the second node device 130. Examples of such data include: any computer program used to operate on the second node device 130.

[0332] The method disclosed in the above embodiment of the present application can be applied to the second processor 132, or implemented by the second processor 132. The second processor 132 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the second processor 132. The above-mentioned second processor 132 may be a general-purpose processor, a digital data processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The second processor 132 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present application, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in the second memory 133. The second processor 132 reads the information in the second memory 133 and completes the steps of the above method in combination with its hardware.

[0333] In an exemplary embodiment, the first node device 120 and the second node device 130 may be implemented by one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD), field programmable gate array (FPGA), general processor, controller, microcontroller (MCU), microprocessor, or other electronic components to execute the aforementioned method.

[0334] It can be understood that the memory (first memory 123, second memory 133) of the embodiment of the present application can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM).The memories described in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memories.

[0335] In an exemplary embodiment, the embodiment of the present invention further provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, for example, including a first memory 123 storing a computer program, and the computer program can be executed by the first processor 122 of the first node device 120 to complete the steps of the aforementioned control server side method. The computer-readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface storage, optical disk, or CD-ROM.

[0336] It should be noted that: "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0337] In addition, the technical solutions described in the embodiments of the present invention can be arbitrarily combined without conflict.

[0338] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. An information processing method, characterized in that: Applied to the first node, the method comprises: Receiving first information respectively sent by at least two second nodes; the first information represents relevant parameters of the machine learning model trained by the second node; For each second node, based on the first information, determining whether the performance of the machine learning model trained by the corresponding second node meets a preset condition; When it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset condition, determining a first learning mode corresponding to the corresponding second node, and notifying the corresponding second node of the first learning mode; Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

2. The method according to claim 1, characterized in that The first information includes a second learning mode of the second node training the machine learning model; and the determining the first learning mode corresponding to the corresponding second node includes: For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode; When it is determined that the second learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain a plurality of nodes to be processed; Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result; Determine the machine learning model with the best performance evaluation in the evaluation results; use the transfer learning mode corresponding to the machine learning model with the best performance evaluation as the first learning mode corresponding to the corresponding second node; The transfer learning mode characterizes that the parameters of the machine learning model are obtained through transfer learning.

3. The method according to claim 2, characterized in that The notifying the corresponding second node of the first learning mode includes: Generate second information; the second information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model; The second information is sent to the corresponding second node.

4. The method according to claim 1, characterized in that: The first information includes a second learning mode of the second node training the machine learning model; and the determining the first learning mode corresponding to the corresponding second node includes: For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the second learning modes of the machine learning models trained by the other nodes are not transfer learning modes; When it is determined that the learning modes of the machine learning models trained by the other nodes are not transfer learning modes, the autonomous learning mode is used as the first learning mode corresponding to the corresponding second node; Among them, the autonomous learning mode represents that the corresponding second node keeps the model parameters unchanged when retraining the machine learning model.

5. The method according to claim 4, characterized in that The notifying the corresponding second node of the first learning mode includes: generating third information; the third information including the name or number of the corresponding second node and the recommended first learning mode; The third information is sent to the corresponding second node.

6. The method according to claim 1, characterized in that The first information includes a second learning mode of the second node training the machine learning model; and the determining the first learning mode corresponding to the corresponding second node includes: For other nodes among the at least two second nodes except the corresponding second node, using the first information, determine whether the learning mode of the machine learning model trained by the other nodes is a transfer learning mode; When it is determined that the learning mode of the machine learning model trained by the other nodes is a transfer learning mode, the other nodes are used as nodes to be processed to obtain multiple nodes to be processed; Evaluate the performance of the machine learning models trained on the multiple nodes to be processed to obtain an evaluation result; The combination of the transfer learning mode and the autonomous learning mode corresponding to the machine learning model whose performance evaluation in the evaluation result is greater than the performance threshold is used as the first learning mode corresponding to the corresponding second node; Among them, the combination of the transfer learning mode and the autonomous learning mode represents that the corresponding second node retrains the machine learning model while keeping the model parameters unchanged and while updating the model parameters.

7. The method according to claim 6, characterized in that The sending to the corresponding second node comprises: Generate fourth information; the fourth information includes the name or number of the corresponding second node, the recommended first learning mode, the migrated source machine learning model, and the name or number of the node corresponding to the migrated source machine learning model; The fourth information is sent to the corresponding second node.

8. An information processing method, characterized in that: Applied to the second node, the method comprises: Infer the predicted value output by the trained machine learning model to obtain the inference result; Evaluating the inference result to obtain an evaluation result; Determining whether to report the evaluation result to the first node; When it is determined to report the evaluation result to the first node, first information is generated; the first information represents relevant parameters of the machine learning model trained by the second node; The first information is sent to the first node.

9. The method according to claim 8, characterized in that The generating of the first information comprises: The first information is generated based on the name or number of the second node, and the learning mode and parameters of the machine learning model trained by the second node.

10. The method according to claim 8, characterized in that The method further comprises: Receiving a first learning mode sent by the first node; the first learning mode is determined by the first node when it is determined that the performance of the machine learning model trained by the second node does not meet a preset condition; The machine learning model is retrained using the first learning mode so that the performance of the machine learning model meets preset conditions.

11. The method according to claim 10, characterized in that The method further comprises: Notifying the third node of the first learning mode; Among them, the first learning mode is used for the third node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

12. An information processing device, characterized in that: include: A receiving unit, configured to receive first information respectively sent by at least two second nodes; The first information represents relevant parameters of the machine learning model trained by the second node; A first processing unit is used to determine, for each second node, based on the first information, whether the performance of the machine learning model trained by the corresponding second node meets a preset condition; and when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset condition, determining a first learning mode corresponding to the corresponding second node, and notifying the corresponding second node of the first learning mode; Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

13. An information processing device, characterized in that: include: A second processing unit is used to infer the predicted value output by the trained machine learning model to obtain an inference result; Evaluating the inference result to obtain an evaluation result; A third processing unit, used to determine whether to report the evaluation result to the first node; When it is determined to report the evaluation result to the first node, generating first information; The first information represents relevant parameters of the machine learning model trained by the second node; A sending unit is used to send the first information to the first node.

14. A first node device, characterized in that: include: A first communication interface, used to receive first information respectively sent by at least two second nodes; The first information represents relevant parameters of the machine learning model trained by the second node; A first processor is used to determine, for each second node, based on the first information, whether the performance of the machine learning model trained by the corresponding second node meets a preset condition; and when it is determined that the performance of the machine learning model trained by the corresponding second node does not meet the preset condition, determining a first learning mode corresponding to the corresponding second node, and notifying the corresponding second node of the first learning mode; Among them, the first learning mode is used for the corresponding second node to retrain the machine learning model so that the performance of the machine learning model meets preset conditions.

15. A second node device, characterized in that: include: A second processor is used to infer the predicted value output by the trained machine learning model to obtain an inference result; Evaluating the inference result to obtain an evaluation result; and determining whether to report the evaluation result to the first node; When it is determined to report the evaluation result to the first node, generating first information; The first information represents relevant parameters of the machine learning model trained by the second node; The second communication interface is used to send the first information to the first node.

16. A first node device, characterized in that: comprising a processor and a memory for storing a computer program capable of being executed on the processor, Wherein, when the processor is used to run the computer program, it executes the steps of the method described in any one of claims 1 to 7.

17. A second node device, characterized in that: comprising a processor and a memory for storing a computer program capable of being executed on the processor, Wherein, when the processor is used to run the computer program, it executes the steps of the method described in any one of claims 8 to 11.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7, or executes the steps of the method according to any one of claims 8 to 11.

Images