A method, device and equipment for generating a multi-device collaboration model

Abstract

The present invention provides a method, device and equipment for generating a multi-device collaboration model, and relates to the field of communication technology. The multi-device collaboration model generation method includes: when a target device is found to join an Internet of Things network, obtaining information provided by the target device; and updating the current collaboration model according to the information provided by the target device to obtain a target collaboration model. In the solution of the present invention, the collaboration of multiple devices is limited to a fixed model, which reduces the flexibility of collaboration of devices in the Internet of Things network.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method, device and equipment for generating a multi-device collaboration model. Background Art

[0002] As smart devices become more and more popular, the average person will own dozens of devices in the future. Multi-device collaboration will greatly increase the number of tasks that can be executed by IoT applications and improve the quality of task completion.

[0003] However, in actual IoT scenarios, there is great uncertainty in the status of devices. Common problems such as equipment failure, human intervention, and scenario changes will affect the effective operation of the collaborative model, causing the model to perform poorly or even fail completely. This impact becomes more significant as the number of devices involved in the collaboration increases. Summary of the invention

[0004] The purpose of the present invention is to provide a method, apparatus and device for generating a multi-device collaboration model, so as to solve the problem that multi-device collaboration is limited to a fixed model, which reduces the flexibility of collaboration among IoT network devices.

[0005] To achieve the above object, an embodiment of the present invention provides a method for generating a multi-device collaboration model, comprising:

[0006] When a target device is found to have joined the Internet of Things network, obtaining information provided by the target device;

[0007] According to the information provided by the target device, the current collaboration model is updated to obtain a target collaboration model.

[0008] Optionally, the information provided by the target device includes at least one of the following:

[0009] Data information;

[0010] Decision-making models;

[0011] Decision results;

[0012] The data information includes original data and / or abstract features.

[0013] Optionally, updating the current collaboration model according to the information provided by the target device to obtain a target collaboration model includes:

[0014] Determine the network structure and model parameters of the target collaboration model according to the current collaboration model and the decision model of the target device;

[0015] Obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model, and the data information and decision result of the target device;

[0016] The target collaboration model is trained according to the network structure, model parameters and the second training sample of the target collaboration model.

[0017] Optionally, obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model, and the data information and decision result of the target device includes:

[0018] Determining a training data label according to a decision result in the first training sample and a decision result of the target device;

[0019] According to the training data label, obtaining a label fusion result and a soft label fusion result of the decision result in the first training sample and the decision result of the target device;

[0020] All data information and the corresponding decision results, label fusion results and soft label fusion results are recorded to obtain the second training sample.

[0021] Optionally, obtaining, according to the training data label, a label fusion result of the decision result in the first training sample and the decision result of the target device and a soft label fusion result includes:

[0022] By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i , the maximum P i The corresponding decision result is taken as the label fusion result R; where Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models;

[0023] By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

[0024] Optionally, the training of the target collaboration model according to the network structure, model parameters and the second training sample of the target collaboration model includes:

[0025] Obtaining distribution information of the second training sample;

[0026] Obtaining a reliability evaluation result of the second training sample;

[0027] Determining an initial value of a training sample weight according to the distribution information and the reliability evaluation result;

[0028] Based on the initial weights of the training samples, the target collaboration model is trained through the second training samples.

[0029] Optionally, the training the target collaboration model by using the second training sample includes:

[0030] Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

[0031] To achieve the above object, an embodiment of the present invention provides a multi-device collaboration model generation apparatus, comprising:

[0032] An acquisition module, used to acquire information provided by the target device when it is found that the target device has joined the Internet of Things network;

[0033] The processing module is used to update the current collaboration model according to the information provided by the target device to obtain a target collaboration model.

[0034] Optionally, the information provided by the target device includes at least one of the following:

[0035] Data information;

[0036] Decision-making models;

[0037] Decision results;

[0038] The data information includes original data and / or abstract features.

[0039] Optionally, the processing module includes:

[0040] A first processing submodule, configured to determine a network structure and model parameters of a target collaboration model according to a current collaboration model and a decision model of the target device;

[0041] A second processing submodule, configured to obtain a second training sample of a target collaboration model according to the first training sample of the current collaboration model and the data information and decision result of the target device;

[0042] The third processing submodule is used to train the target collaboration model according to the network structure, model parameters and the second training sample of the target collaboration model.

[0043] Optionally, the second processing submodule includes:

[0044] A first processing unit, configured to determine a training data label according to a decision result in the first training sample and a decision result of the target device;

[0045] A second processing unit is used to obtain a label fusion result and a soft label fusion result of the decision result in the first training sample and the decision result of the target device according to the training data label;

[0046] The third processing unit is used to record all data information and corresponding decision results, label fusion results and soft label fusion results to obtain a second training sample.

[0047] Optionally, the second processing unit is further used for:

[0048] By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i , the maximum P i The corresponding decision result is taken as the label fusion result R; where Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models;

[0049] By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

[0050] Optionally, the third processing submodule includes:

[0051] A fourth processing unit, configured to obtain distribution information of the second training sample;

[0052] A fifth processing unit, configured to obtain a reliability evaluation result of the second training sample;

[0053] A sixth processing unit, configured to determine an initial value of a training sample weight according to the distribution information and the reliability evaluation result;

[0054] A seventh processing unit is used to train the target collaboration model through the second training samples based on the initial weights of the training samples.

[0055] Optionally, the seventh processing unit is further configured to:

[0056] Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

[0057] To achieve the above object, an embodiment of the present invention provides a network side device, including: a transceiver and a processor;

[0058] The transceiver is used to obtain information provided by the target device when it is found that the target device has joined the Internet of Things network;

[0059] The processor is used to update the current collaboration model according to the information provided by the target device to obtain a target collaboration model.

[0060] Optionally, the information provided by the target device includes at least one of the following:

[0061] Data information;

[0062] Decision-making models;

[0063] Decision results;

[0064] The data information includes original data and / or abstract features.

[0065] Optionally, the processor is further configured to:

[0066] Determine the network structure and model parameters of the target collaboration model according to the current collaboration model and the decision model of the target device;

[0067] Obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model, and the data information and decision result of the target device;

[0068] The target collaboration model is trained according to the network structure, model parameters and the second training sample of the target collaboration model.

[0069] Optionally, the processor is further configured to:

[0070] Determining a training data label according to a decision result in the first training sample and a decision result of the target device;

[0071] According to the training data label, obtaining a label fusion result and a soft label fusion result of the decision result in the first training sample and the decision result of the target device;

[0072] All data information and the corresponding decision results, label fusion results and soft label fusion results are recorded to obtain the second training sample.

[0073] Optionally, the processor is further configured to:

[0074] By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th devicei , the maximum P i The corresponding decision result is taken as the label fusion result R; where Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models;

[0075] By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

[0076] Optionally, the processor is further configured to:

[0077] Obtaining distribution information of the second training sample;

[0078] Obtaining a reliability evaluation result of the second training sample;

[0079] Determining an initial value of a training sample weight according to the distribution information and the reliability evaluation result;

[0080] Based on the initial weights of the training samples, the target collaboration model is trained through the second training samples.

[0081] Optionally, the processor is further configured to:

[0082] Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

[0083] To achieve the above-mentioned purpose, an embodiment of the present invention provides a mobile terminal, including a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, the multi-device collaboration model generation method as described above is implemented.

[0084] To achieve the above objectives, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the multi-device collaboration model generation method as described above are implemented.

[0085] The beneficial effects of the above technical solution of the present invention are as follows:

[0086] The method of the embodiment of the present invention, when discovering that the target device has joined the Internet of Things network, will obtain the information provided by the target device, thereby updating the current collaboration model with the information provided by the target device, and further obtaining a target collaboration model for multi-device collaboration after the target device joins, so as to avoid multi-device collaboration being limited to a fixed model and enhance the flexibility of collaboration among devices in the Internet of Things network. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A flowchart of a method for generating a multi-device collaboration model according to an embodiment of the present invention;

[0088] Figure 2 Construct a schematic diagram for the network structure of the target collaboration model;

[0089] Figure 3 Construct a schematic diagram for the network structure of the target collaboration model;

[0090] Figure 4 Generate a schematic diagram for the training samples of the target collaborative model;

[0091] Figure 5 A structural diagram of a multi-device collaboration model generating apparatus according to an embodiment of the present invention;

[0092] Figure 6 is a structural diagram of a network side device according to an embodiment of the present invention;

[0093] Figure 7 This is a structural diagram of a network side device according to another embodiment of the present invention. DETAILED DESCRIPTION

[0094] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0095] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present invention. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0096] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0097] Additionally, the terms "system" and "network" are often used interchangeably herein.

[0098] In the embodiments provided in the present application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0099] like Figure 1 As shown, a method for generating a multi-device collaboration model according to an embodiment of the present invention includes:

[0100] Step 101, when it is found that a target device has joined the Internet of Things network, obtaining information provided by the target device;

[0101] Step 102: Update the current collaboration model according to the information provided by the target device to obtain a target collaboration model.

[0102] The method of the embodiment of the present invention, through steps 101 and 102, will obtain the information provided by the target device when it is found that the target device joins the Internet of Things network, so as to update the current collaboration model with the information provided by the target device, and further obtain a target collaboration model for multi-device collaboration after the target device joins, so as to avoid the multi-device collaboration being limited to a fixed model and improve the flexibility of the collaboration of the Internet of Things network devices.

[0103] Among them, the collaboration model can also be a decision model, and the current collaboration model as the update object is the collaboration model of multiple devices in the Internet of Things network. In this way, the method of the embodiment of the present invention can be applied to the scenario where a new device joins the Internet of Things network, and the new device is used as the target device to update the collaboration model. For example, in the Internet of Vehicles business scenario, decisions are made jointly for the information of devices near each intersection. By applying the method of the embodiment of the present invention, it is possible to achieve rapid collaboration of networked devices, automatically build a decision model that conforms to reality, improve the decision accuracy and fault tolerance, and improve the humanization and practicality of the product. Or, in the smart home scenario, decisions are made jointly for the information of household devices such as cameras, speakers, infrared sensors, etc. By applying the method of the embodiment of the present invention, the generated collaboration model can achieve higher collaboration accuracy, and when some products fail, the remaining devices can still meet some needs, thereby improving the service quality and user satisfaction of the product. Or, in the scenario of using multiple wearable devices, decisions are made jointly for each device. By applying the method of the embodiment of the present invention, the generated collaboration model can improve the health monitoring capability, and the user will not be aware of the device collaboration mode and the replacement of the collaboration model during use, thereby improving the user's product experience.

[0104] Optionally, the information provided by the target device includes at least one of the following:

[0105] Data information;

[0106] Decision-making models;

[0107] Decision results;

[0108] The data information includes original data and / or abstract features.

[0109] Here, the original data can be the data collected by the target device itself, or the data collected by other devices received by the target device; the abstract features are the data features obtained after analyzing and processing the original data, and the specific abstract feature content is set according to the device's own capabilities. The decision model is a single model that the target device itself has, and accordingly, the decision result is obtained by the decision model.

[0110] In this embodiment, optionally, step 102 includes:

[0111] Determine the network structure and model parameters of the target collaboration model according to the current collaboration model and the decision model of the target device;

[0112] Obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model, and the data information and decision result of the target device;

[0113] The target collaboration model is trained according to the network structure, model parameters and the second training sample of the target collaboration model.

[0114] Here, the current collaboration model and the decision model of the newly added target device are combined to jointly determine the network structure and model parameters of the target collaboration model; and the second training sample of the target collaboration model is further obtained from the first training sample of the current collaboration model, the data information of the target device and the decision result of the target device; finally, based on the second training sample, the target collaboration model with the determined network structure and model parameters is trained to obtain the required target collaboration model to meet the collaboration requirements of multiple devices in the IoT network after the target device joins the IoT network.

[0115] If the target device has no decision-making capability, the network structure and model parameters of the target collaboration model can retain the network structure and model parameters of the current collaboration model, and the target collaboration model can be obtained by training by using the data information of the target device to expand the first training sample, such as Figure 2 As shown; if the target device has decision-making ability, in addition to using the data information of the target device to expand the first training sample, the current collaboration model and the decision model of the newly added target device can be retained to guide the learning of the target collaboration model, and the target collaboration model is obtained by training, such as Figure 3 Specifically, when the network structure of the target collaboration model is being constructed, invalid connections in the model will be removed and target device connections will be added.

[0116] In this embodiment, the method of the embodiment of the present invention generates a target collaboration model that can retain the network structure and model parameters of the original model (i.e., the collaboration model before the target device joins the Internet of Things network), and add the original data and / or abstract features provided by the target device to the input space (i.e., the storage space of the model input).

[0117] In addition, in the embodiment of the present invention, optionally, obtaining a second training sample of the target collaboration model according to the first training sample of the current collaboration model and the data information and decision result of the target device includes:

[0118] Determining a training data label according to a decision result in the first training sample and a decision result of the target device;

[0119] According to the training data labels, respectively obtaining label fusion results and soft label fusion results of the decision results in the first training sample and the decision results of the target device;

[0120] All data information and the corresponding decision results, label fusion results and soft label fusion results are recorded to obtain the second training sample.

[0121] Here, firstly, the training data label is determined by the decision result in the first training sample and the decision result of the target device; then, based on the determined training data label, the decision fusion result of the decision result in the first training sample and the decision result of the target device is obtained, that is, the label fusion result and the soft label fusion result; then, all data information and the corresponding decision results, label fusion results and soft label fusion results are recorded to obtain the second training sample. For example, the second training sample is Figure 4 As shown. In this way, in addition to data information and decision results, the second training sample also records the label fusion results and soft label fusion results, so that subsequent training through the second training sample can also use intermediate parameters to assist training and improve training efficiency.

[0122] The training data label can be understood as the class (or category) of the training data, and the setting of the label (i.e. hard label) is based on the device capabilities. For example, for health monitoring of multiple wearable devices, the training data labels can be divided into "sitting", "standing", "exercise" and other categories according to the user's behavior. The soft label refers to the probability vector formed by the softmax layer, and each value in the vector corresponds to the probability that the model recognizes the sample as a certain class.

[0123] It should be noted that only some devices with decision-making capabilities can provide soft tag information.

[0124] Based on the determined training data label, optionally, obtaining a label fusion result and a soft label fusion result of a decision result in the first training sample and a decision result of the target device according to the training data label includes:

[0125] By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i , the maximum P i The corresponding decision result is taken as the label fusion result R; where Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models;

[0126] By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

[0127] Here, after the target device joins the IoT network, the N devices in the IoT network are calculated by the formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i After that, the maximum P can be selected from the reputation voting values ​​of N devices. i The corresponding decision result is taken as the label fusion result R. For the soft label fusion result p, the formula p = ∑ i p i ×Acc i Calculate. Where i represents the i-th device among N devices.

[0128] In addition, considering that different samples in the training samples have different effects on model training, in this embodiment, optionally, training the target collaboration model according to the network structure, model parameters and the second training sample of the target collaboration model includes:

[0129] Obtaining distribution information of the second training sample;

[0130] Obtaining a reliability evaluation result of the second training sample;

[0131] Determining an initial value of a training sample weight according to the distribution information and the reliability evaluation result;

[0132] Based on the initial weights of the training samples, the target collaboration model is trained through the second training samples.

[0133] In this way, through the distribution information and reliability evaluation results of the second training sample, we can understand the guidance of the sample to the model, thereby determining the optimal initial value of the training sample weight, and finally training the target collaborative model through the second training sample based on the determined initial weight of the training sample.

[0134] The distribution information reflects the distribution of the data information (i.e., the sample) in the second training sample in the input space. Specifically, the distribution information Q s =q2-q1, q1 is the same category space evaluation parameter, q2 is the different category space evaluation parameter. q1 can be calculated by calculating the distance of each sample from the class center C of the same class j. j The distance d j Afterwards, by d j Divide by the average distance D of class j j The result is q1 = d j / D j q2 can be obtained by calculating the distance of each sample from the class center C of other classes k. k The distance d k Afterwards, by d k Divide by the category C k The average distance D k , take the minimum value among all the results obtained, that is, q2 = mind k / D k .

[0135] The reliability evaluation result reflects the direction of the data information (i.e., sample) in the second training sample for model training. Here, the reliability of the training sample is evaluated based on the label fusion result and the soft label fusion result. The consistent decision results of multiple devices reflect that the sample is more reliable. Therefore, specifically, the reliability evaluation result Q r =s'+λp', λ is the balance parameter, s' is the information entropy of s, and p' is the information entropy of the soft label fusion result p. Among them, s is the credibility voting value P of all decision results i Accumulate them according to categories and get vectors corresponding to all categories.

[0136] After obtaining the distribution information Q s And the reliability evaluation result Q r Finally, the sample difficulty evaluation Q can be obtained by adding the two and normalizing them, thereby determining that the initial value of the training sample weight is equal to Q.

[0137] After determining the initial value of the training sample weight, optionally, training the target collaboration model by using the second training sample includes:

[0138] Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

[0139] Here, model distillation means that the intermediate parameters of the original model can also assist in the training of the new model. In this embodiment, the target collaborative model is trained simultaneously using soft labels and labels. In this way, the loss function used in the training is:

[0140]

[0141] Wherein g is the g-th data information in the second training sample, and m is the total number of data information in the second training sample. The softmax layer output of the target collaborative model corresponding to the g-th data information is compared with the soft label fusion result corresponding to the g-th data information to obtain the soft label error; is the category output of the target collaborative model corresponding to the g-th data information, and is compared with the label fusion result corresponding to the g-th data information to obtain the label error. g is the training sample weight, at the beginning of training v g =Q, then the weights of the training samples gradually balance until v g =1.

[0142] To summarize, the method of the embodiment of the present invention, when discovering that the target device has joined the Internet of Things network, will obtain the information provided by the target device, thereby updating the current collaboration model with the information provided by the target device, and further obtaining a target collaboration model for multi-device collaboration after the target device has joined, so as to avoid the multi-device collaboration being limited to a fixed model and enhance the flexibility of collaboration among devices in the Internet of Things network.

[0143] like Figure 5 As shown, a multi-device collaboration model generation apparatus according to an embodiment of the present invention includes:

[0144] An acquisition module 510 is used to acquire information provided by the target device when it is found that the target device has joined the Internet of Things network;

[0145] The processing module 520 is used to update the current collaboration model according to the information provided by the target device to obtain a target collaboration model.

[0146] Optionally, the information provided by the target device includes at least one of the following:

[0147] Data information;

[0148] Decision-making models;

[0149] Decision results;

[0150] The data information includes original data and / or abstract features.

[0151] Optionally, the processing module includes:

[0152] A first processing submodule, configured to determine a network structure and model parameters of a target collaboration model according to a current collaboration model and a decision model of the target device;

[0153] A second processing submodule, configured to obtain a second training sample of a target collaboration model according to the first training sample of the current collaboration model and the data information and decision result of the target device;

[0154] The third processing submodule is used to train the target collaboration model according to the network structure, model parameters and the second training sample of the target collaboration model.

[0155] Optionally, the second processing submodule includes:

[0156] A first processing unit, configured to determine a training data label according to a decision result in the first training sample and a decision result of the target device;

[0157] A second processing unit is used to obtain a label fusion result and a soft label fusion result of the decision result in the first training sample and the decision result of the target device according to the training data label;

[0158] The third processing unit is used to record all data information and corresponding decision results, label fusion results and soft label fusion results to obtain a second training sample.

[0159] Optionally, the second processing unit is further used for:

[0160] By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i , the maximum P i The corresponding decision result is taken as the label fusion result R; where Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models;

[0161] By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

[0162] Optionally, the third processing submodule includes:

[0163] A fourth processing unit, configured to obtain distribution information of the second training sample;

[0164] A fifth processing unit, configured to obtain a reliability evaluation result of the second training sample;

[0165] A sixth processing unit, configured to determine an initial value of a training sample weight according to the distribution information and the reliability evaluation result;

[0166] A seventh processing unit is used to train the target collaboration model through the second training samples based on the initial weights of the training samples.

[0167] Optionally, the seventh processing unit is further configured to:

[0168] Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

[0169] In this way, the apparatus of an embodiment of the present invention, when discovering that a target device has joined the Internet of Things network, will obtain the information provided by the target device, thereby updating the current collaboration model with the information provided by the target device, and further obtaining a target collaboration model for collaboration of multiple devices after the target device has joined, so as to avoid limiting the collaboration of multiple devices to a fixed model and enhance the flexibility of collaboration of devices in the Internet of Things network.

[0170] It should be noted that the device is a device that applies the above-mentioned multi-device collaboration model generation method, and the implementation method of the above-mentioned method embodiment is applicable to the device and can also achieve the same technical effect.

[0171] like Figure 6 As shown, a network side device according to an embodiment of the present invention includes: a transceiver 610 and a processor 620;

[0172] The transceiver 610 is used to obtain information provided by the target device when it is found that the target device has joined the Internet of Things network;

[0173] The processor 620 is used to update the current collaboration model according to the information provided by the target device to obtain a target collaboration model.

[0174] Optionally, the information provided by the target device includes at least one of the following:

[0175] Data information;

[0176] Decision-making models;

[0177] Decision results;

[0178] The data information includes original data and / or abstract features.

[0179] Optionally, the processor is further configured to:

[0180] Determine the network structure and model parameters of the target collaboration model according to the current collaboration model and the decision model of the target device;

[0181] Obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model, and the data information and decision result of the target device;

[0182] The target collaboration model is trained according to the network structure, model parameters and the second training sample of the target collaboration model.

[0183] Optionally, the processor is further configured to:

[0184] Determining a training data label according to a decision result in the first training sample and a decision result of the target device;

[0185] According to the training data label, obtaining a label fusion result and a soft label fusion result of the decision result in the first training sample and the decision result of the target device;

[0186] All data information and the corresponding decision results, label fusion results and soft label fusion results are recorded to obtain the second training sample.

[0187] Optionally, the processor is further configured to:

[0188] By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i , the maximum P i The corresponding decision result is taken as the label fusion result R; where Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models;

[0189] By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

[0190] Optionally, the processor is further configured to:

[0191] Obtaining distribution information of the second training sample;

[0192] Obtaining a reliability evaluation result of the second training sample;

[0193] Determining an initial value of a training sample weight according to the distribution information and the reliability evaluation result;

[0194] Based on the initial weights of the training samples, the target collaboration model is trained through the second training samples.

[0195] Optionally, the processor is further configured to:

[0196] Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

[0197] The network side device of this embodiment, when discovering that the target device has joined the Internet of Things network, will obtain the information provided by the target device, thereby updating the current collaboration model with the information provided by the target device, and further obtaining a target collaboration model for multi-device collaboration after the target device joins, so as to avoid multi-device collaboration being limited to a fixed model and enhance the flexibility of collaboration among Internet of Things network devices.

[0198] A network side device according to another embodiment of the present invention, such as Figure 7 As shown, it includes a transceiver 710, a processor 700, a memory 720, and a computer program stored in the memory 720 and executable on the processor 700; when the processor 700 executes the computer program, the above-mentioned multi-device collaboration model generation method is implemented.

[0199] The transceiver 710 is used to receive and send data under the control of the processor 700.

[0200] Among them, Figure 7 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 700 and memory represented by memory 720. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver 710 may be a plurality of components, i.e., including a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 700 is responsible for managing the bus architecture and general processing, and the memory 720 may store data used by the processor 700 when performing operations.

[0201] A computer-readable storage medium according to an embodiment of the present invention stores a computer program, which, when executed by a processor, implements the steps in the multi-device collaboration model generation method described above, and can achieve the same technical effect. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0202] It should be further explained that many functional components described in this specification are referred to as modules in order to more particularly emphasize the independence of their implementation methods.

[0203] In the embodiment of the present invention, module can be implemented with software so that it can be executed by various types of processors. For example, an executable code module of an identification can include one or more physical or logical blocks of computer instructions, for example, it can be constructed as an object, process or function. Nevertheless, the executable code of the identified module does not need to be physically located together, but can include different instructions stored in different positions, and when these instructions are logically combined together, it constitutes a module and realizes the specified purpose of the module.

[0204] In fact, executable code module can be a single instruction or many instructions, and can even be distributed on a plurality of different code segments, distributed among different programs, and distributed across a plurality of memory devices. Similarly, operating data can be identified in the module, and can be implemented and organized in the data structure of any appropriate type according to any appropriate form. The operating data can be collected as a single data set, or can be distributed in different locations (including on different storage devices), and can only be present on a system or network as an electronic signal at least in part.

[0205] When a module can be implemented by software, considering the level of existing hardware technology, a person skilled in the art can build a corresponding hardware circuit to implement the corresponding function of the module that can be implemented by software without considering the cost. The hardware circuit includes a conventional very large scale integration (VLSI) circuit or gate array and existing semiconductors such as logic chips, transistors, or other discrete components. The module can also be implemented by a programmable hardware device, such as a field programmable gate array, a programmable array logic, a programmable logic device, etc.

[0206] The above exemplary embodiments are described with reference to the accompanying drawings, and many different forms and embodiments are feasible without departing from the spirit and teachings of the present invention. Therefore, the present invention should not be constructed as a limitation of the exemplary embodiments proposed herein. More specifically, these exemplary embodiments are provided so that the present invention will be perfect and complete, and the scope of the present invention will be conveyed to those who are familiar with the technology. In these figures, the component sizes and relative sizes may be exaggerated for clarity. The terms used here are only based on the purpose of describing specific exemplary embodiments and are not intended to be limiting. As used herein, unless the text clearly indicates otherwise, the singular forms "one", "an" and "the" are intended to include these multiple forms. It will be further understood that the terms "including" and / or "comprising" when used in this specification indicate the presence of the features, integers, steps, operations, components and / or components, but do not exclude the presence or increase of one or more other features, integers, steps, operations, components, components and / or their groups. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of that range and any subranges therebetween.

[0207] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for generating a multi-device collaboration model, characterized in that: include: When a target device is found to have joined the Internet of Things network, obtaining information provided by the target device; According to the information provided by the target device, the current collaboration model is updated to obtain a target collaboration model; The information provided by the target device includes at least one of the following: Data information; Decision-making models; Decision results; Wherein, the data information includes original data and / or abstract features; The updating of the current collaboration model according to the information provided by the target device to obtain the target collaboration model includes: Determine the network structure and model parameters of the target collaboration model according to the current collaboration model and the decision model of the target device; Obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model, and the data information and decision result of the target device; Training the target collaboration model according to the network structure, model parameters and the second training sample of the target collaboration model; The step of obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model and the data information and decision result of the target device includes: Determining a training data label according to a decision result in the first training sample and a decision result of the target device; Obtaining, according to the training data label, a label fusion result of the decision result in the first training sample and the decision result of the target device and a soft label fusion result; All data information and the corresponding decision results, label fusion results and soft label fusion results are recorded to obtain the second training sample.

2. The method according to claim 1, characterized in that The step of obtaining, according to the training data label, a label fusion result of the decision result in the first training sample and the decision result of the target device and a soft label fusion result includes: By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i , the maximum P i The corresponding decision result is taken as the label fusion result R; where Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models; By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

3. The method according to claim 1, characterized in that The step of training the target collaboration model according to the network structure, model parameters and second training sample of the target collaboration model includes: Obtaining distribution information of the second training sample; Obtaining a reliability evaluation result of the second training sample; Determining an initial value of a training sample weight according to the distribution information and the reliability evaluation result; Based on the initial value of the training sample weight, the target collaboration model is trained through the second training sample.

4. The method according to claim 3, characterized in that The step of training the target collaboration model by using the second training sample includes: Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

5. A multi-device collaboration model generation device, characterized in that: include An acquisition module, used to acquire information provided by the target device when it is found that the target device has joined the Internet of Things network; A processing module, used for updating the current collaboration model according to the information provided by the target device to obtain a target collaboration model; The information provided by the target device includes at least one of the following: Data information; Decision-making models; Decision results; Wherein, the data information includes original data and / or abstract features; Wherein, the processing module includes: A first processing submodule, configured to determine a network structure and model parameters of a target collaboration model according to a current collaboration model and a decision model of the target device; A second processing submodule, configured to obtain a second training sample of a target collaboration model according to the first training sample of the current collaboration model and the data information and decision result of the target device; A third processing submodule, configured to train the target collaboration model according to the network structure, model parameters and second training sample of the target collaboration model; The second processing submodule includes: A first processing unit, configured to determine a training data label according to a decision result in the first training sample and a decision result of the target device; A second processing unit is used to obtain a label fusion result and a soft label fusion result of the decision result in the first training sample and the decision result of the target device according to the training data label; The third processing unit is used to record all data information and corresponding decision results, label fusion results and soft label fusion results to obtain a second training sample.

6. The device according to claim 5, characterized in that The second processing unit is further configured to: By formula P i =R i ×Acc i , calculate the reputation voting value P of the i-th device i , the maximum P i The corresponding decision result is taken as the label fusion result R; Among them, Acc i represents the accuracy of a single decision model for the i-th device, R i Indicates the consistency between the single decision model of the i-th device and other decision models; By formula p = ∑ i p i ×Acc i , calculate the soft label fusion result p; where p i Represents the decision soft label of a single decision model for the i-th device.

7. The device according to claim 5, characterized in that The third processing submodule includes: A fourth processing unit, configured to obtain distribution information of the second training sample; A fifth processing unit, configured to obtain a reliability evaluation result of the second training sample; A sixth processing unit, configured to determine an initial value of a training sample weight according to the distribution information and the reliability evaluation result; A seventh processing unit is used to train the target collaboration model through the second training sample based on the initial value of the training sample weight.

8. The device according to claim 7, characterized in that The seventh processing unit is further configured to: Model distillation is adopted to train the target collaborative model using the labels and soft labels of the data information in the second training sample.

9. A network side device, characterized in that: include: transceivers and processors; The transceiver is used to obtain information provided by the target device when it is found that the target device has joined the Internet of Things network; The processor is used to update the current collaboration model according to the information provided by the target device to obtain a target collaboration model; The information provided by the target device includes at least one of the following: Data information; Decision-making models; Decision results; Wherein, the data information includes original data and / or abstract features; Wherein, the processor is further used for: Determine the network structure and model parameters of the target collaboration model according to the current collaboration model and the decision model of the target device; Obtaining a second training sample of a target collaboration model according to the first training sample of the current collaboration model, and the data information and decision result of the target device; Training the target collaboration model according to the network structure, model parameters and the second training sample of the target collaboration model; The processor is further configured to: Determining a training data label according to a decision result in the first training sample and a decision result of the target device; Obtaining, according to the training data label, a label fusion result of the decision result in the first training sample and the decision result of the target device and a soft label fusion result; All data information and the corresponding decision results, label fusion results and soft label fusion results are recorded to obtain the second training sample.

10. A network side device, comprising: A transceiver, a processor, a memory, and a computer program stored in the memory and executable on the processor; characterized in that when the processor executes the computer program, the multi-device collaboration model generation method as described in any one of claims 1 to 4 is implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the multi-device collaboration model generation method as described in any one of claims 1 to 4 are implemented.

Images