Model transmission method, model transmission device and storage medium

Abstract

The invention relates to a model transmission method, a model transmission device and a storage medium. The model transmission method is applied to a network device, and the method comprises the following steps: in response to receiving at least one model distribution request, determining a first model, and segmenting the first model to obtain a first number of model segmentation blocks; wherein each model distribution request in the at least one model distribution request corresponds to a model segmentation block. According to the method and the device, the model can be selected according to the model distribution request, and the segmented model segmentation blocks are segmented and sent, so that the reasoning and training capabilities of the terminal on the model can be reduced, and communication resources can be reduced.

Description

[0001] The present disclosure is a divisional application of application number 202080004148.6, and the name of the invention is: A model transmission method, a model transmission device and a storage medium. Technical Field

[0002] The present disclosure relates to the field of wireless communication technology, and in particular to a model transmission method, a model transmission device and a storage medium. Background Art

[0003] In order to meet the needs of multiple business scenarios, communication networks must have the characteristics of ultra-high speed, ultra-low latency, ultra-high reliability and ultra-large number of connections. These business scenarios, as well as the corresponding needs and characteristics of communication networks, have brought unprecedented challenges to the deployment and operation and maintenance of communication networks.

[0004] In related technologies, artificial intelligence is introduced to improve the resource utilization of communication networks, terminal service experience, and automation and intelligent control and management of communication networks. The model obtained through deep learning of artificial intelligence can have better performance. However, when sending the model to the terminal, if the entire model is sent to the terminal, not only is the amount of data transmitted relatively large, but also for the terminal, its computing power is difficult to meet the needs of model reasoning and training. Summary of the invention

[0005] In order to overcome the problems existing in the related art, the present disclosure provides a model transmission method, a model transmission device and a storage medium.

[0006] According to a first aspect of an embodiment of the present disclosure, a model transmission method is provided, which is applied to a network device, and the method includes:

[0007] In response to receiving at least one model distribution request, a first model is determined, and the first model is segmented to obtain a first number of model segmentation blocks; wherein each model distribution request in the at least one model distribution request corresponds to a model segmentation block.

[0008] In one embodiment, the model distribution request includes a first model accuracy, a first model reasoning complexity, and a first model data volume;

[0009] In response to receiving at least one model request, determining a first model includes:

[0010] Determine the second model accuracy, second model reasoning complexity and second model data volume of each of the multiple models; based on predefined operation rules, determine the first operation value of the first model accuracy and the second model accuracy, the second operation value of the first model reasoning complexity and the second model reasoning complexity, and the third operation value of the first model data volume and the second model data volume; determine the model corresponding to the minimum sum of the first operation value, the second operation value and the third operation value as the first model.

[0011] In one implementation, the predefined rule is determined in the following manner:

[0012] In response to the second model accuracy being greater than or equal to the first model accuracy, determining that the first operation value is zero as the operation rule;

[0013] or

[0014] In response to the second model accuracy being less than the first model accuracy, determining the first operation value to be the square of the difference between the first model accuracy and the second model accuracy as the operation rule;

[0015] or

[0016] In response to the first model reasoning complexity being greater than or equal to the second model reasoning complexity, determining that the second operation value is zero as the operation rule;

[0017] or

[0018] In response to the first model reasoning complexity being less than the second model reasoning complexity, determining the second operation value as the square of the difference between the first model reasoning complexity and the second model reasoning complexity as the operation rule;

[0019] or

[0020] In response to the first model data amount being greater than or equal to the second model data amount, determining that the third operation value is zero as the operation rule;

[0021] or

[0022] In response to the first model data amount being smaller than the second model data amount, the third operation value is determined as the square of the difference between the first model data amount and the second model data amount as the operation rule.

[0023] In one embodiment, the model distribution request includes computing resources and communication resources, and the method further includes:

[0024] Determine a first model data volume of the first model, and determine a plurality of model segmentation point options based on the first model data volume, computing resources, and communication resources, wherein the model segmentation point options are used to segment the first model.

[0025] In one embodiment, the method further comprises:

[0026] A first indication message is sent, wherein the first indication message includes the plurality of model segmentation point options.

[0027] In one embodiment, the model distribution request includes a first model split point option;

[0028] The segmenting of the first model comprises:

[0029] At least one of said first model split point options is determined.

[0030] In one embodiment, the method further comprises:

[0031] Determine a merging threshold of the model segments, and determine a difference value between every two of the model segments in a first number of model segments; in response to a first difference value in the difference values ​​being less than or equal to the merging threshold, determine that the two model segments corresponding to the first difference value are the same, and merge the model segments to obtain a second number of model segments.

[0032] In one implementation, determining a merge threshold of a model segment block includes:

[0033] Determine a first model reasoning complexity and determine a channel quality; determine a merging threshold proportional to the inverse of the product of the first model reasoning complexity and the channel quality.

[0034] In one embodiment, determining the difference value between every two of the model segments in the first number of model segments includes:

[0035] A matrix of a first number of model partitions is determined, and a difference value between every two of the model partitions is determined based on the matrix.

[0036] In one embodiment, the method further comprises:

[0037] A second indication message is sent, wherein the second indication message includes at least one model segmentation block among the second number of model segmentation blocks.

[0038] In one implementation, the sending the second indication message includes:

[0039] In response to a first model segmentation block in the second number of model segmentation blocks corresponding to a model distribution request, determining to send a second indication message based on unicast;

[0040] or

[0041] In response to a first model segmentation block in the second number of model segmentation blocks corresponding to a plurality of model distribution requests, determining to send a second indication message based on multicast;

[0042] or

[0043] In response to a first model segmentation block in the second number of model segmentation blocks corresponding to all model distribution requests, it is determined to send a second indication message based on broadcasting.

[0044] According to a second aspect of an embodiment of the present disclosure, a model transmission method is provided, which is applied to a terminal, and the method includes:

[0045] Send a model distribution request; wherein the model distribution request corresponds to a model segmentation block.

[0046] In one embodiment, the method further comprises:

[0047] A first indication message is received, wherein the first indication message includes a plurality of model segmentation point options, wherein the model segmentation point options are used to segment a model; and a first model segmentation point option is determined based on the first indication message.

[0048] In one implementation, determining the first model segmentation point option based on the first indication message includes:

[0049] According to the needs of the terminal itself, a third number of model segmentation point options are determined from the multiple model segmentation point options; the first model reasoning complexity and the first model data volume are determined, and the third model reasoning complexity and the third model data volume of the model segmentation block corresponding to each of the model segmentation point options in the third number of model segmentation point options are determined; based on predefined operation rules, fourth operation values ​​of the first model reasoning complexity and the third model reasoning complexity, and fifth operation values ​​of the first model data volume and the third model data volume are determined; the model segmentation point option corresponding to the minimum sum of the fourth operation value and the fifth operation value is determined as the model segmentation point option.

[0050] In one embodiment, the predefined operation rules include:

[0051] In response to the first model reasoning complexity being greater than or equal to the third model reasoning complexity, determining that the fourth operation value is zero as the operation rule;

[0052] or

[0053] In response to the first model reasoning complexity being less than the third model reasoning complexity, determining the fourth operation value to be the square of the difference between the first model reasoning complexity and the third model reasoning complexity as the operation rule;

[0054] or

[0055] In response to the first model data amount being greater than or equal to the third model data amount, determining that the fifth operation value is zero as the operation rule;

[0056] or

[0057] In response to the first model data amount being smaller than the third model data amount, determining the fifth operation value to be the square of the difference between the first model data amount and the third model data amount as the operation rule.

[0058] In one embodiment, the method further comprises:

[0059] A second indication message is received, wherein the second indication message includes at least one model segmentation block of the second number of model segmentation blocks.

[0060] According to a third aspect of an embodiment of the present disclosure, a model transmission device is provided, which is applied to a network device, and the device includes:

[0061] A determination module is used to determine a first model in response to receiving at least one model distribution request, and to segment the first model to obtain a first number of model segmentation blocks; wherein each model distribution request in the at least one model distribution request corresponds to a model segmentation block.

[0062] In one embodiment, the model distribution request includes a first model accuracy, a first model reasoning complexity, and a first model data volume;

[0063] The determination module is used for:

[0064] Determine the second model accuracy, second model reasoning complexity and second model data volume of each of the multiple models; based on predefined operation rules, determine the first operation value of the first model accuracy and the second model accuracy, the second operation value of the first model reasoning complexity and the second model reasoning complexity, and the third operation value of the first model data volume and the second model data volume; determine the model corresponding to the minimum sum of the first operation value, the second operation value and the third operation value as the first model.

[0065] In one implementation, the determining module is used to:

[0066] In response to the second model accuracy being greater than or equal to the first model accuracy, determining that the first operation value is zero as the operation rule;

[0067] or

[0068] In response to the second model accuracy being less than the first model accuracy, determining the first operation value to be the square of the difference between the first model accuracy and the second model accuracy as the operation rule;

[0069] or

[0070] In response to the first model reasoning complexity being greater than or equal to the second model reasoning complexity, determining that the second operation value is zero as the operation rule;

[0071] or

[0072] In response to the first model reasoning complexity being less than the second model reasoning complexity, determining the second operation value as the square of the difference between the first model reasoning complexity and the second model reasoning complexity as the operation rule;

[0073] or

[0074] In response to the first model data amount being greater than or equal to the second model data amount, determining that the third operation value is zero as the operation rule;

[0075] or

[0076] In response to the first model data amount being smaller than the second model data amount, the third operation value is determined as the square of the difference between the first model data amount and the second model data amount as the operation rule.

[0077] In one implementation, the model distribution request includes computing resources and communication resources, and the determination module is further configured to:

[0078] Determine a first model data volume of the first model, and determine a plurality of model segmentation point options based on the first model data volume, computing resources, and communication resources, wherein the model segmentation point options are used to segment the first model.

[0079] In one embodiment, the device further comprises:

[0080] A first indication message is sent, wherein the first indication message includes the plurality of model segmentation point options.

[0081] In one embodiment, the model distribution request includes a first model split point option;

[0082] The determination module is used for:

[0083] At least one of said first model split point options is determined.

[0084] In one embodiment, the determining module is further configured to:

[0085] Determine a merging threshold of the model segments, and determine a difference value between every two of the model segments in a first number of model segments; in response to a first difference value in the difference values ​​being less than or equal to the merging threshold, determine that the two model segments corresponding to the first difference value are the same, and merge the model segments to obtain a second number of model segments.

[0086] In one implementation, the determining module is used to:

[0087] Determine a first model reasoning complexity and determine a channel quality; determine a merging threshold proportional to the inverse of the product of the first model reasoning complexity and the channel quality.

[0088] In one implementation, the determining module is used to:

[0089] A matrix of a first number of model partitions is determined, and a difference value between every two of the model partitions is determined based on the matrix.

[0090] In one embodiment, the device further comprises:

[0091] A second indication message is sent, wherein the second indication message includes at least one model segmentation block among the second number of model segmentation blocks.

[0092] In one implementation, the determining module is used to:

[0093] In response to a first model segmentation block in the second number of model segmentation blocks corresponding to a model distribution request, determining to send a second indication message based on unicast;

[0094] or

[0095] In response to a first model segmentation block in the second number of model segmentation blocks corresponding to a plurality of model distribution requests, determining to send a second indication message based on multicast;

[0096] or

[0097] In response to a first model segmentation block in the second number of model segmentation blocks corresponding to all model distribution requests, it is determined to send a second indication message based on broadcasting.

[0098] According to a fourth aspect of an embodiment of the present disclosure, a model transmission device is provided, which is applied to a terminal, and the device includes:

[0099] The sending module is used to send a model distribution request; wherein the model distribution request corresponds to a model segmentation block.

[0100] In one embodiment, the device further comprises:

[0101] A first indication message is received, wherein the first indication message includes a plurality of model segmentation point options, wherein the model segmentation point options are used to segment a model; and a first model segmentation point option is determined based on the first indication message.

[0102] In one embodiment, the device further comprises:

[0103] According to the needs of the terminal itself, a third number of model segmentation point options are determined from the multiple model segmentation point options; the first model reasoning complexity and the first model data volume are determined, and the third model reasoning complexity and the third model data volume of the model segmentation block corresponding to each of the model segmentation point options in the third number of model segmentation point options are determined; based on predefined operation rules, fourth operation values ​​of the first model reasoning complexity and the third model reasoning complexity, and fifth operation values ​​of the first model data volume and the third model data volume are determined; the model segmentation point option corresponding to the minimum sum of the fourth operation value and the fifth operation value is determined as the model segmentation point option.

[0104] In one embodiment, the predefined operation rules include:

[0105] In response to the first model reasoning complexity being greater than or equal to the third model reasoning complexity, determining that the fourth operation value is zero as the operation rule;

[0106] or

[0107] In response to the first model reasoning complexity being less than the third model reasoning complexity, determining the fourth operation value to be the square of the difference between the first model reasoning complexity and the third model reasoning complexity as the operation rule;

[0108] or

[0109] In response to the first model data amount being greater than or equal to the third model data amount, determining that the fifth operation value is zero as the operation rule;

[0110] or

[0111] In response to the first model data amount being smaller than the third model data amount, determining the fifth operation value to be the square of the difference between the first model data amount and the third model data amount as the operation rule.

[0112] In one embodiment, the device further comprises:

[0113] A second indication message is received, wherein the second indication message includes at least one model segmentation block of the second number of model segmentation blocks.

[0114] According to a fifth aspect of an embodiment of the present disclosure, there is provided a model transmission device, including:

[0115] A processor; a memory for storing processor executable instructions; wherein the processor is configured to: execute the model transmission method described in the first aspect or any one of the embodiments of the first aspect, or execute the model transmission method described in the second aspect or any one of the embodiments of the second aspect.

[0116] According to the sixth aspect of an embodiment of the present disclosure, a non-temporary computer-readable storage medium is provided. When the instructions in the storage medium are executed by a processor of a mobile terminal, the mobile terminal is enabled to execute the model transmission method described in the first aspect or any one of the embodiments of the first aspect, or the mobile terminal is enabled to execute the model transmission method described in the second aspect or any one of the embodiments of the second aspect.

[0117] The technical solution provided by the embodiments of the present disclosure may include the following beneficial effects: by receiving a model distribution request, determining a first model among multiple models, and segmenting the first model, the selected model can achieve relatively high model accuracy and relatively low model complexity in all terminals, and model segmentation can reduce communication overhead in subsequent transmissions.

[0118] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0119] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0120] Figure 1 A schematic diagram of the system architecture of a model transmission method provided in the present invention.

[0121] Figure 2 The figure is a flow chart of a model transmission method according to an exemplary embodiment.

[0122] Figure 3 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0123] Figure 4 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0124] Figure 5 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0125] Figure 6 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0126] Figure 7 The figure is a flowchart of a model transmission method according to an exemplary embodiment.

[0127] Figure 8 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0128] Fig. 9 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0129] Fig.10 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0130] Fig.11 is a flowchart of yet another model transmission method according to an exemplary embodiment.

[0131] Fig.12 The figure is a flowchart of a model transmission method according to an exemplary embodiment.

[0132] Fig.13 It is a protocol and interface principle diagram of the model selection and model segmentation parts in a model transmission method according to an exemplary embodiment.

[0133] Fig.14 It is a protocol and interface principle diagram of the multicast / broadcast / unicast transmission and distribution part of a model segmentation block in a model transmission method according to an exemplary embodiment.

[0134] Fig.15 The figure is a block diagram of a model transmission device according to an exemplary embodiment.

[0135] Fig.16 is a block diagram of yet another model transmission device according to an exemplary embodiment.

[0136] Fig.17 The invention is a block diagram of a device for model transmission according to an exemplary embodiment.

[0137] Fig.18 is a block diagram of yet another device for model transmission according to an exemplary embodiment. DETAILED DESCRIPTION

[0138] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0139] In order to meet the needs of multiple business scenarios, communication networks must have the characteristics of ultra-high speed, ultra-low latency, ultra-high reliability and ultra-large number of connections. These business scenarios, as well as the corresponding needs and characteristics of communication networks, have brought unprecedented challenges to the deployment and operation and maintenance of communication networks.

[0140] With the breakthrough of artificial intelligence technology, especially in the enrichment of deep learning algorithms, the improvement of hardware computing power, and the introduction of massive data in the new generation of communication networks, it provides strong support for the intelligence of the new generation of networks. In addition, artificial intelligence can be used to further improve the resource utilization of communication networks, improve the terminal service experience of communication networks, and realize the automation and intelligent control and management of communication networks.

[0141] In related technologies, deep learning has achieved unprecedented performance in processing many tasks. By obtaining multiple models based on deep learning, better performance can be achieved. The terminal obtains the required model by sending a model distribution request to the network side. Generally, the process of the terminal applying for a model from the network side can be carried out in the following ways:

[0142] The terminal initiates a model distribution request to the network side (e.g., base station), the base station sends a measurement configuration request to the user, the terminal determines the measurement result (e.g., channel quality) according to the radio resource control (RRC) measurement configuration, and the terminal sends the determined measurement result to the base station. The base station determines the transmission resource allocation according to the channel measurement result, and sends the model to the terminal in a unicast transmission mode according to the transmission resource allocation result. The terminal executes the reception model according to the allocated resources and determines the model.

[0143] According to the process of the terminal applying for a model from the network side in the above embodiment, it can be determined that the related technology has the following deficiencies:

[0144] (1) If the base station does not make a reasonable choice of the model, the model distributed to the terminal may not be able to achieve relatively high model accuracy and model complexity on all terminals. It may also cause the problem of excessive transmission load, resulting in a shortage of wireless resources.

[0145] (2) The base station distributes the model directly to all terminals. Therefore, the amount of data used to transmit the model between the base station and the terminals is very large, which will also cause the problem of excessive transmission load and increase communication overhead.

[0146] (3) After the terminal determines the model, it needs to infer and train the model. If the terminal obtains the entire model, high requirements are placed on the terminal's computing power during inference and training. Therefore, there will be a part of the terminal whose computing power cannot meet the requirements of model inference and training.

[0147] In view of the above shortcomings, the present disclosure provides a model transmission method, which selects the model required by the terminal, segments the model, and distributes the segmented model blocks to the terminal. This can not only reduce unnecessary communication overhead and improve the utilization rate of wireless network resources, but also reduce the terminal's requirements for model reasoning and training, solving the problem of excessive data volume of deep learning models and the difficulty of terminal computing power to meet computing needs.

[0148] Figure 1 A schematic diagram of the system architecture of a model transmission method provided by the present disclosure. Figure 1 As shown, the system includes a core network part and a wireless access network part. The terminal (user) accesses the base station through a wireless channel, and the base stations are connected through the Xn interface. The base station accesses the User Port Function (UPF) network element of the core network through the N3 interface. The UPF network element accesses the Session Management Function (SMF) network element through the N4 interface. The SMF network element accesses the bus structure of the core network and is connected to other network functions (NF) of the core network.

[0149] Understandably, Figure 1 The communication system of the network device and the terminal shown is only for schematic illustration. The wireless communication system may also include other network devices, such as wireless relay devices and wireless backhaul devices. Figure 1 The embodiment of the present disclosure does not limit the number of network devices and terminals included in the wireless communication system.

[0150] It can be further understood that the wireless communication system of the embodiment of the present disclosure is a network that provides wireless communication functions. The wireless communication system can adopt different communication technologies, such as code division multiple access (CDMA), wideband code division multiple access (WCDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal frequency-division multiple access (OFDMA), single carrier frequency division multiple access (SC-FDMA), carrier sense multiple access / collision avoidance (Carrier Sense Multiple Access with Collision Avoidance). According to the capacity, rate, delay and other factors of different networks, the network can be divided into 2G (English: generation) network, 3G network, 4G network or future evolution network, such as 5G network, 5G network can also be called new radio network (New Radio, NR). For the convenience of description, the present disclosure sometimes refers to the wireless communication network as a network.

[0151] Furthermore, the network equipment involved in the present disclosure may also be referred to as a wireless access network device. The wireless access network device may be: a base station, an evolved node B (base station), a home base station, an access point (AP) in a wireless fidelity (WIFI) system, a wireless relay node, a wireless backhaul node, a transmission point (TP) or a transmission and reception point (TRP), etc. It may also be a gNB in ​​an NR system, or it may also be a component or part of a device constituting a base station. When it is a vehicle-to-everything (V2X) communication system, the network device may also be a vehicle-mounted device. It should be understood that in the embodiments of the present disclosure, the specific technology and specific device form adopted by the network device are not limited.

[0152] Furthermore, the terminal involved in the present disclosure may also be referred to as a terminal device, a user equipment (User Equipment, UE), a mobile station (Mobile Station, MS), a mobile terminal (Mobile Terminal, MT), etc., which is a device that provides voice and / or data connectivity to users. For example, the terminal may be a handheld device with a wireless connection function, a vehicle-mounted device, etc. At present, some examples of terminals are: a smart phone (Mobile Phone), a pocket computer (Pocket Personal Computer, PPC), a PDA, a personal digital assistant (Personal Digital Assistant, PDA), a laptop computer, a tablet computer, a wearable device, or a vehicle-mounted device, etc. In addition, when it is a vehicle-to-everything (V2X) communication system, the terminal device may also be a vehicle-mounted device. It should be understood that the embodiments of the present disclosure do not limit the specific technology and specific device form adopted by the terminal.

[0153] Figure 2 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Figure 2 As shown, the model transmission method is used in a network device and includes the following steps.

[0154] In step S11 , in response to receiving at least one model distribution request, a first model is determined, and the first model is segmented to obtain a first number of model segmentation blocks.

[0155] In the embodiment of the present disclosure, each model distribution request in at least one model distribution request corresponds to a model segmentation block.

[0156] In an embodiment of the present disclosure, a network device (eg, a base station) receives a model distribution request initiated by at least one terminal, wherein the model distribution request may also include information such as model accuracy and generalization performance of the distribution model.

[0157] The base station determines to send RRC configuration (RRCReconfiguration) to the terminal according to the received model distribution request. The RRCReconfiguration is sent to obtain information such as measurement configuration, mobility control, and radio resource configuration. The base station also receives the measurement result sent by the terminal through a measurement report message (MeasurementReport).

[0158] The base station performs joint model selection among multiple local models for all received model distribution requests according to at least one received model distribution request, determines a first model for distribution to each terminal, divides the first model, determines a first number of model division blocks, and sends the model division blocks to the corresponding terminal based on the received model distribution request.

[0159] In the disclosed embodiment, by performing joint model selection on the model distribution request, the first model for each terminal is determined, and the first model is segmented, and the model segmentation blocks are sent to the corresponding terminals. This can not only flexibly adapt the model segmentation blocks required by each terminal, but also solve the problem that the terminal computing power is difficult to meet the reasoning and training model.

[0160] The model transmission method provided in the present disclosure includes determining a first model in response to receiving at least one model request.

[0161] Figure 3 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Figure 3 As shown, in response to receiving at least one model request, determining a first model includes the following steps.

[0162] In step S21, the second model accuracy, the second model reasoning complexity and the second model data volume of each of the multiple models are determined.

[0163] In an embodiment of the present disclosure, the model distribution request received by the base station may also include the terminal's requirements for model accuracy, model reasoning complexity, and model data volume. For ease of distinction, the present disclosure refers to the terminal's requirements for model accuracy, model reasoning complexity, and model data volume as first model accuracy, first model reasoning complexity, and first model data volume. The base station determines the model accuracy, model reasoning complexity, and model data volume of each model based on the models in the local area. Similarly, for ease of distinction, the present disclosure refers to the model accuracy, model reasoning complexity, and model data volume corresponding to each model included in the base station as second model accuracy, second model reasoning complexity, and second model data volume.

[0164] In step S22, based on predefined operation rules, first operation values ​​of the first model accuracy and the second model accuracy, second operation values ​​of the first model reasoning complexity and the second model reasoning complexity, and third operation values ​​of the first model data volume and the second model data volume are determined.

[0165] In the embodiment of the present disclosure, the predefined operation rule may be determined by the following formula:

[0166]

[0167] In the formula, x represents the unknown number and y represents the unknown number.

[0168] Exemplarily, the first model accuracy and the second model accuracy are substituted into the above formula to determine the first operation value, in which case x can be the second model accuracy and y can be the first model accuracy. The first model reasoning complexity and the second model reasoning complexity are substituted into the above formula to determine the second operation value, in which case x can be the second model reasoning complexity and y can be the first model reasoning complexity. The first model data volume and the second model data volume are substituted into the above formula to determine the third operation value, in which case x can be the second model data volume and y can be the first model data volume.

[0169] In step S23, the model corresponding to the minimum sum of the first operation value, the second operation value and the third operation value is determined as the first model.

[0170] In the embodiment of the present disclosure, the minimum value of the sum of the first operation value, the second operation value, and the third operation value may be determined by the following formula:

[0171]

[0172] In the formula, a u is the first model accuracy, b u is the first model reasoning complexity, c u is the data volume of the first model, m is the first model; a m is the second model accuracy, b m is the inference complexity of the second model, c m is the amount of data for the second model. U is the number of model distribution requests.

[0173] In the embodiment of the present disclosure, the predefined operation rules are determined in the following manner:

[0174] In response to the second model accuracy being greater than or equal to the first model accuracy, determining the first operation value to be zero as an operation rule. Or, in response to the second model accuracy being less than the first model accuracy, determining the first operation value to be the square of the difference between the first model accuracy and the second model accuracy as an operation rule. Or, in response to the first model reasoning complexity being greater than or equal to the second model reasoning complexity, determining the second operation value to be zero as an operation rule. Or, in response to the first model reasoning complexity being less than the second model reasoning complexity, determining the second operation value to be the square of the difference between the first model reasoning complexity and the second model reasoning complexity as an operation rule. Or, in response to the first model data volume being greater than or equal to the second model data volume, determining the third operation value to be zero as an operation rule. Or, in response to the first model data volume being less than the second model data volume, determining the third operation value to be the square of the difference between the first model data volume and the second model data volume as an operation rule.

[0175] In an embodiment of the present disclosure, the base station determines the computing power resources and communication resources of the terminal included in the received model distribution request. The first model data volume of the first model is determined, and based on the computing power resources and communication resources required to be consumed by each layer of the model in the first model, and model characteristics such as the first model data volume, multiple model segmentation point options are determined. Among them, the model segmentation point option is used to segment the first model.

[0176] In an embodiment of the present disclosure, the base station may send a first indication message to indicate the determined multiple model segmentation point options.

[0177] In an embodiment of the present disclosure, the model distribution request received by the base station also includes a first model segmentation point option. The base station determines to segment the first model according to at least one first model segmentation point option received, and divides the first model into different model segmentation blocks, that is, obtains a first number of model segmentation blocks. The base station can also merge the model segmentation blocks that meet the merging conditions for the first number of model segmentation blocks to obtain a second number of model segmentation blocks, wherein it can be understood that the first number and the second number can be the same, or the second number can be less than the first number.

[0178] In the embodiment of the present disclosure, the base station merges the first number of model segmentation blocks according to the merging condition to obtain the second number of model segmentation blocks, and the following implementation methods may be adopted.

[0179] Figure 4 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Figure 4 As shown, the model transmission method is applied to a network device and also includes the following steps.

[0180] In step S31 , a merging threshold of the model segments is determined, and a difference value between every two model segments in the first number of model segments is determined.

[0181] In step S32, in response to a first difference value in the difference values ​​being less than or equal to a merging threshold, it is determined that two model segments corresponding to the first difference value are the same, and the model segments are merged to obtain a second number of model segments.

[0182] In the embodiment of the present disclosure, the base station compares all the determined difference values ​​with the merging threshold. The average value of the two model segmentation blocks corresponding to the first difference value less than or equal to the merging threshold can be determined, and the two model segmentation blocks are merged according to the average value. In the embodiment of the present disclosure, the model segmentation blocks are merged to obtain the second number of model segmentation blocks using the following formula:

[0183]

[0184] Where m = 1, ..., M represents the index of different model segmentation blocks, Pm represents the parameter matrix of the mth model segmentation block, and P represents the parameter matrix of the merged model segmentation block.

[0185] The parameter matrix of the merged model segmentation blocks is obtained, that is, the second number of model segmentation blocks is obtained.

[0186] In the embodiment of the present disclosure, the model transmission method further includes determining a merging threshold of the model segmentation blocks.

[0187] Figure 5 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Figure 5 As shown, determining the merging threshold of the model segmentation blocks includes the following steps.

[0188] In step S41, the first model inference complexity is determined, and the channel quality is determined.

[0189] In step S42, a merging threshold is determined to be proportional to the inverse of the product of the first model inference complexity and the channel quality.

[0190] In the embodiment of the present disclosure, the base station determines the first model reasoning complexity and the channel quality, and the merging threshold is proportional to the inverse of the product of the first model reasoning complexity and the channel quality, which can be expressed by the following formula:

[0191]

[0192] In the formula, a is the first model reasoning complexity, d is the value of channel quality, ε Indicates the merge threshold for determining whether to merge model segments.

[0193] Figure 6 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Figure 6 As shown, determining the difference value between every two model segments in the first number of model segments includes the following steps.

[0194] In step S51 , matrices of a first number of model segments are determined, and a difference value between every two model segments is determined based on the matrices.

[0195] In the embodiment of the present disclosure, the base station may determine the matrix of the first number of model partition blocks, such as the parameter matrix of the model partition blocks. The difference values ​​between the model partition blocks are determined by the parameter matrix of the model partition blocks, and each difference value is compared with the size of the merging threshold. The method may adopt the following formula:

[0196]

[0197] In the formula, is the parameter matrix of the model segmentation block, m For the m Model partition blocks, n For the n Model partition blocks, ε The merging threshold is used to determine whether to perform the model segmentation block merging operation, where I is the number of rows in the parameter matrix, J is the number of columns in the parameter matrix, i is the i-th row in the parameter matrix, and j is the j-th row in the parameter matrix.

[0198] In some embodiments of the present disclosure, according to the above formula, in response to a first difference value in the difference value being less than or equal to the merging threshold, the base station determines a second difference value corresponding to the first difference value. m The model partition block and n In response to the first difference value in the difference value being greater than the merge threshold, determining the m The model partition block and n The model segmentation blocks are different.

[0199] In the embodiment of the present disclosure, the base station determines a second number of model segmentation blocks and sends a second indication message, wherein the second indication message includes at least one model segmentation block among the second number of model segmentation blocks.

[0200] In the disclosed embodiment, the base station determines the model distribution request corresponding to each model segmentation block and determines the sending mode of sending the second indication message, wherein the sending mode of the base station sending the second indication message includes unicast mode, multicast mode and broadcast mode.

[0201] In one embodiment of the present disclosure, in response to a first model segmentation block in the second number of model segmentation blocks corresponding to a model distribution request, it is determined that the base station sends a second indication message to a terminal corresponding to the first segmentation model block based on unicast.

[0202] In one embodiment of the present disclosure, in response to the first model segmentation block in the second number of model segmentation blocks corresponding to multiple model distribution requests, it is determined to send a second indication message based on multicast. The base station requests the Multimedia Broadcast / Multicast Service (MBMS) GW to multicast the model segmentation block; the MBMS GW sends an MBMS authorization request to the BM-SC to seek authorization for the terminal to receive data, and requests the user to activate the MBMS context; the terminal sends an activation MBMS context request to the MBMS GW, and then the MBMS GW sends an MBMS registration request to the BM-SC to obtain the MBMS bearer information of the MBMS bearer service, which is used to verify whether the MBMS bearer capability of the terminal meets the required MBMS bearer capability.

[0203] In one embodiment of the present disclosure, in response to the first model segmentation block in the second number of model segmentation blocks corresponding to all model distribution requests, it is determined to send the second indication message based on broadcast. The base station requests the MBMS GW to broadcast the model segmentation block; the MBMS GW sends an MBMS authorization request to the BM-SC to seek authorization for the terminal to receive data; and verifies whether the MBMS bearer capability of the terminal meets the required MBMS bearer capability.

[0204] Figure 7 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Figure 7 As shown, the method comprises the following steps:

[0205] Step S61: The terminal initiates a model distribution request to the base station.

[0206] Step S62: The base station performs joint model selection for all terminals according to the inference task requirements and communication overhead of the terminals.

[0207] Step S63: the base station determines model segmentation point options for the model according to the model features, and sends the model segmentation point selection options to the terminal.

[0208] Step S64: The terminal determines the model segmentation point options according to the local computing capability and communication conditions, and reports the model segmentation point options to the base station.

[0209] Step S65: the base station merges the model segmentation blocks according to the selection of the model segmentation points by the terminal, and sends the merged model segmentation blocks to the terminal according to the corresponding transmission mode.

[0210] Step S66: After the terminal obtains the model segmentation blocks, the model distribution process ends.

[0211] Based on the same / similar concept, the embodiment of the present disclosure also provides a model transmission method.

[0212] Figure 8 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Figure 8 As shown, the model transmission method is applied to the terminal, including the following steps.

[0213] In step S71, a model distribution request is sent.

[0214] In the disclosed embodiment, the model distribution request corresponds to the model segmentation block. The terminal can request the corresponding model distribution request from the network device according to its own needs. Among them, the model distribution request also includes model requirements for reasoning tasks and communication overhead, which may include requirements for model accuracy, model reasoning complexity and model data volume.

[0215] Fig. 9 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Fig. 9 As shown, the model transmission method is applied to the terminal and also includes the following steps.

[0216] In step S81, a first indication message is received.

[0217] In an embodiment of the present disclosure, the first indication message includes a plurality of model segmentation point options, wherein the model segmentation point options are used to segment the model.

[0218] In step S82, a first model segmentation point option is determined based on the first indication message.

[0219] In an embodiment of the present disclosure, the terminal filters the received model segmentation point options according to its own needs and capabilities (for example, computing power, reasoning, and model training capabilities).

[0220] Fig.10 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Fig.10 As shown, determining a first model segmentation point option based on a first indication message includes the following steps.

[0221] In step S91 , according to the needs of the terminal itself, a third number of model segmentation point options are determined from a plurality of model segmentation point options.

[0222] In step S92, the first model reasoning complexity and the first model data volume are determined, and the third model reasoning complexity and the third model data volume of the model segmentation block corresponding to each model segmentation point option in the third number of model segmentation point options are determined.

[0223] In step S93, based on predefined operation rules, fourth operation values ​​of the first model reasoning complexity and the third model reasoning complexity, and fifth operation values ​​of the first model data volume and the third model data volume are determined.

[0224] In step S94, the model segmentation point option corresponding to the minimum sum of the fourth operation value and the fifth operation value is determined as the first model segmentation point option.

[0225] In the embodiment of the present disclosure, the terminal determines a third number of model segmentation point options from the multiple model segmentation point options according to the received multiple model segmentation point options and the needs and capabilities of the terminal itself. For example, the third number of model segmentation point options is N model segmentation point options. Taking the nth as an example, the terminal determines its own first model reasoning complexity and first model data volume, and determines the third model reasoning complexity and third model data volume of the model segmentation block corresponding to the nth model segmentation point option.

[0226] In some embodiments of the present disclosure, the terminal determines the fourth operation value of the first model reasoning complexity and the third model reasoning complexity, and the fifth operation value of the first model data volume and the third model data volume according to a predefined operation rule. The predefined operation rule can be determined using the following formula.

[0227]

[0228] In the formula, x represents the unknown number and y represents the unknown number.

[0229] Exemplarily, the fourth operation value can be determined by substituting the first model reasoning complexity and the third model reasoning complexity into the above formula, in which case x can be the first model reasoning complexity and y can be the third model reasoning complexity. The fifth operation value can be determined by substituting the first model data volume and the third model data volume into the above formula, in which case x can be the first model data volume and y can be the third model data volume.

[0230] In the embodiment of the present disclosure, the terminal determines the model segmentation point option corresponding to the minimum value according to the determined fourth operation value and the fifth operation value, and determines the model segmentation point option as the first model segmentation point option. The first model segmentation point option can also be determined by the following formula:

[0231]

[0232] Where n is the nth model split point option, a is the first model reasoning complexity, b is the first model data volume, and a n is the third model reasoning complexity of the model segmentation block corresponding to the nth model segmentation point option, b n The third model data volume of the model segmentation block corresponding to the nth model segmentation point option.

[0233] In some embodiments of the present disclosure, the predefined operation rules may be determined in the following manner:

[0234] In response to the first model reasoning complexity being greater than or equal to the third model reasoning complexity, the fourth operation value is determined to be zero as an operation rule. Or, in response to the first model reasoning complexity being less than the third model reasoning complexity, the fourth operation value is determined to be the square of the difference between the first model reasoning complexity and the third model reasoning complexity as an operation rule. Or, in response to the first model data volume being greater than or equal to the third model data volume, the fifth operation value is determined to be zero as an operation rule. Or, in response to the first model data volume being less than the third model data volume, the fifth operation value is determined to be the square of the difference between the first model data volume and the third model data volume as an operation rule.

[0235] Fig.11FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Fig.11 As shown, the model transmission method is applied to the terminal and also includes the following steps.

[0236] In step S101, a second indication message is received.

[0237] In an embodiment of the present disclosure, the second indication message includes at least one model segmentation block among the second number of model segmentation blocks.

[0238] Fig.12 FIG. 1 is a flow chart showing a model transmission method according to an exemplary embodiment. Fig.12 As shown, the method comprises the following steps:

[0239] In step S111, the base station divides the first model into different model segmentation blocks according to the selection of the model segmentation point options by the terminal.

[0240] In step S112, the base station determines whether the model segment blocks meet the merging condition, and merges the model segment blocks that meet the condition.

[0241] In step S113, the base station determines a transmission mode for each merged model segment block.

[0242] In step S114, the base station allocates transmission resources to all terminals according to the transmission mode of each model segment block.

[0243] In step S115, the base station sends the corresponding model segmentation blocks to the terminal using the corresponding transmission mode according to the transmission resource allocation result.

[0244] Fig.13 FIG. 1 is a protocol and interface schematic diagram of a model selection and model segmentation part in a model transmission method according to an exemplary embodiment. Fig.13 As shown, it mainly involves the terminal, base station, and AI service module in the model transmission method provided in the embodiment of the present disclosure. Here, the AI ​​service module is actually contained in the base station, which is responsible for providing the deep learning model and related parameters; the base station is responsible for data transmission and control signaling interaction with the terminal, as well as the selection, segmentation and resource allocation of the first model, and its implementation method is as follows:

[0245] The terminal sends a model distribution request signaling to the base station, and the signaling indicates the content: initiating a model distribution request to the base station. The base station sends the model distribution request signaling to the AI ​​service module. The AI ​​service module sends a model structure and hyperparameter information signaling to the base station, and the signaling indicates the content: sending the model structure and hyperparameter information to the recipient. The base station sends an RRCReconfiguration signaling for sending measurement configuration to the terminal, and the signaling indicates the content: sending the RRCReconfiguration for measurement configuration to the terminal. According to the RRC measurement configuration, the terminal sends a MeasurementReport feedback measurement result signaling to the base station, and the signaling indicates the content: sending the MeasurementReport measurement result to the base station. The terminal sends a model requirement signaling for sending inference tasks and communication overhead to the base station, and the signaling indicates the content: sending the model requirement for inference tasks and communication overhead to the recipient. The base station performs joint model selection for all terminals in the existing model space based on the inference task requirements and communication overhead of the terminal. The base station sends a model distribution response signaling to the terminal, and the signaling indicates the content: sending a model distribution response to the terminal. The base station determines the model segmentation point options for the model based on the model characteristics. The terminal sends a signaling of sending a model segmentation point option to the terminal, and the signaling indicates that the model segmentation point option is sent to the receiver. The terminal determines the model segmentation point according to the local computing capability and communication conditions. The terminal sends a signaling of sending a model segmentation point to the base station, and the signaling indicates that the model segmentation point is sent to the receiver.

[0246] Fig.14 FIG. 1 is a schematic diagram of the protocol and interface of the multicast / broadcast / unicast transmission and distribution part of the model segmentation block in a model transmission method according to an exemplary embodiment. Fig.14 As shown, it mainly involves the terminal, base station, AI service module, MBMS GW, and BM-SC in the model transmission method provided in the embodiment of the present disclosure. Among them, MBMS GW provides an interface for entities using MBMS bearers; BM-SC provides configuration and delivery functions for MBMS user services. Its implementation method is as follows:

[0247] The base station determines the transmission mode of the model segmentation block according to the model segmentation point options of all terminals. If the transmission mode of the model segmentation block is unicast transmission: the base station sends a unicast transmission request signaling to the terminal, and the signaling indicates the content: a unicast transmission request is initiated to the terminal. The terminal sends a unicast transmission response signaling to the base station, and the signaling indicates the content: notifying the base station to accept the unicast transmission model segmentation block. If the transmission mode of the model segmentation block is multicast transmission: the base station sends a multicast transmission request signaling to the MBMS GW, and the signaling indicates the content: a multicast transmission request is initiated to the MBMS GW. Apply for MBMS multicast transmission service between the terminal, the base station, the AI ​​service module, the MBMS GW and the BM-SC. This includes the MBMS authorization process between the BM-SC and the MBMS GW, the MBMS context activation process between the MBMS GW and the terminal, the MBMS registration process between the BM-SC and the MBMS GW, and the process of the MBMS GW verifying the terminal's MBMS bearer capability. The MBMS GW sends a multicast transmission response signaling to the base station, and the signaling indicates the content: notifying the base station to start multicast transmission of the model segmentation block. If the transmission mode of the model segmentation block is broadcast transmission: the base station sends a broadcast transmission request signaling to the MBMS GW, and the signaling indicates the content: initiating a broadcast transmission request to the MBMS GW. Apply for MBMS broadcast transmission service between the MBMS GW and the BM-SC. This includes the MBMS authorization process between the BM-SC and the MBMS GW and the process of the MBMS GW verifying the MBMS bearer capability of the terminal. The MBMS GW sends a broadcast transmission response signaling to the base station, and the signaling indicates the content: notifying the base station to start broadcasting the model segmentation block. The base station sends a model transmission response signaling to the AI ​​service module, and the signaling indicates the content: notifying the AI ​​service module to start model transmission. The AI ​​service module sends the corresponding model segmentation block to the base station. The base station sends a model transmission request signaling to the terminal, and the signaling indicates the content: initiating a model transmission request to the terminal. An RRC connection is established between the base station and the terminal. The base station allocates resources for the transmission of each model segmentation block. The base station sends a PDCCH resource allocation signaling to the terminal, and the signaling indicates the content: allocating PDCCH resources to the receiver. The base station sends a PDSCH resource allocation signaling to the terminal, and the signaling indicates that the PDSCH resource is allocated to the receiver. The base station sends the corresponding model segmentation block to the terminal. The terminal performs reception of the multicast transmission model segmentation block according to the allocated resources.

[0248] Based on the same concept, the embodiment of the present disclosure also provides a model transmission device.

[0249] It is understandable that the model transmission device provided by the embodiment of the present disclosure includes hardware structures and / or software modules corresponding to the execution of each function in order to realize the above functions. In combination with the units and algorithm steps of each example disclosed in the embodiment of the present disclosure, the embodiment of the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the technical solution of the embodiment of the present disclosure.

[0250] Fig.15 is a block diagram of a model transmission device according to an exemplary embodiment. Fig.15 The model transmission device 100 is applied to a network device and includes a determination module 101.

[0251] The determination module 101 is configured to determine a first model in response to receiving at least one model distribution request, and to segment the first model to obtain a first number of model segmentation blocks, wherein each model distribution request in the at least one model distribution request corresponds to a model segmentation block.

[0252] In an embodiment of the present disclosure, the model distribution request includes the first model accuracy, the first model reasoning complexity, and the first model data volume.

[0253] The determination module 101 is used to determine the second model accuracy, the second model reasoning complexity, and the second model data volume of each of the multiple models. Based on the predefined operation rules, the first operation value of the first model accuracy and the second model accuracy, the second operation value of the first model reasoning complexity and the second model reasoning complexity, and the third operation value of the first model data volume and the second model data volume are determined. The model corresponding to the minimum sum of the first operation value, the second operation value, and the third operation value is determined as the first model.

[0254] In the embodiment of the present disclosure, the determination module 101 is used to, in response to the second model accuracy being greater than or equal to the first model accuracy, determine that the first operation value is zero as an operation rule. Or, in response to the second model accuracy being less than the first model accuracy, determine that the first operation value is the square of the difference between the first model accuracy and the second model accuracy as an operation rule. Or, in response to the first model reasoning complexity being greater than or equal to the second model reasoning complexity, determine that the second operation value is zero as an operation rule. Or, in response to the first model reasoning complexity being less than the second model reasoning complexity, determine that the second operation value is the square of the difference between the first model reasoning complexity and the second model reasoning complexity as an operation rule. Or, in response to the first model data volume being greater than or equal to the second model data volume, determine that the third operation value is zero as an operation rule. Or, in response to the first model data volume being less than the second model data volume, determine that the third operation value is the square of the difference between the first model data volume and the second model data volume as an operation rule.

[0255] In an embodiment of the present disclosure, the model distribution request includes computing power resources and communication resources. The determination module 101 is also used to determine the first model data volume of the first model, and determine multiple model segmentation point options based on the first model data volume, computing power resources and communication resources, wherein the model segmentation point option is used to segment the first model.

[0256] In the embodiment of the present disclosure, the device further includes:

[0257] A first indication message is sent, where the first indication message includes a plurality of model segmentation point options.

[0258] In the embodiment of the present disclosure, the model distribution request includes a first model segmentation point option.

[0259] The determination module 101 is used to determine at least one first model segmentation point option.

[0260] In the embodiment of the present disclosure, the determination module is further used to determine a merging threshold of the model segments, and determine the difference value between every two model segments in the first number of model segments. In response to a first difference value in the difference value being less than or equal to the merging threshold, it is determined that the two model segments corresponding to the first difference value are the same, and the model segments are merged to obtain the second number of model segments.

[0261] In the embodiment of the present disclosure, the determination module 101 is used to determine the first model reasoning complexity and determine the channel quality. The merging threshold is determined to be proportional to the inverse of the product of the first model reasoning complexity and the channel quality.

[0262] In the embodiment of the present disclosure, the determination module 101 is used to determine matrices of a first number of model partition blocks, and determine the difference value between every two model partition blocks based on the matrix.

[0263] In the embodiment of the present disclosure, the device further includes:

[0264] A second indication message is sent, where the second indication message includes at least one model segmentation block among the second number of model segmentation blocks.

[0265] In the embodiment of the present disclosure, the determination module 101 is used to determine that the second indication message is sent based on unicast in response to the first model segmentation block in the second number of model segmentation blocks corresponding to one model distribution request. Or, in response to the first model segmentation block in the second number of model segmentation blocks corresponding to multiple model distribution requests, determine that the second indication message is sent based on multicast. Or, in response to the first model segmentation block in the second number of model segmentation blocks corresponding to all model distribution requests, determine that the second indication message is sent based on broadcast.

[0266] Fig.16 is a block diagram of a model transmission device according to an exemplary embodiment. Figure 6 The model transmission device 200 is applied to a terminal and includes a sending module 201.

[0267] The sending module 201 is used to send a model distribution request, wherein the model distribution request corresponds to a model segmentation block.

[0268] In the embodiment of the present disclosure, the device further includes:

[0269] A first indication message is received, wherein the first indication message includes a plurality of model segmentation point options, wherein the model segmentation point options are used to segment the model, and a first model segmentation point option is determined based on the first indication message.

[0270] In the embodiment of the present disclosure, the device further includes:

[0271] According to the needs of the terminal itself, a third number of model segmentation point options are determined from multiple model segmentation point options. The first model reasoning complexity and the first model data volume are determined, and the third model reasoning complexity and the third model data volume of the model segmentation block corresponding to each model segmentation point option in the third number of model segmentation point options are determined. Based on predefined operation rules, the fourth operation value of the first model reasoning complexity and the third model reasoning complexity, and the fifth operation value of the first model data volume and the third model data volume are determined. The model segmentation point option corresponding to the minimum sum between the fourth operation value and the fifth operation value is determined as the model segmentation point option.

[0272] In the embodiment of the present disclosure, the predefined operation rules include:

[0273] In response to the first model reasoning complexity being greater than or equal to the third model reasoning complexity, the fourth operation value is determined to be zero as an operation rule. Or, in response to the first model reasoning complexity being less than the third model reasoning complexity, the fourth operation value is determined to be the square of the difference between the first model reasoning complexity and the third model reasoning complexity as an operation rule. Or, in response to the first model data volume being greater than or equal to the third model data volume, the fifth operation value is determined to be zero as an operation rule. Or, in response to the first model data volume being less than the third model data volume, the fifth operation value is determined to be the square of the difference between the first model data volume and the third model data volume as an operation rule.

[0274] In the embodiment of the present disclosure, the device further includes:

[0275] A second indication message is received, the second indication message including at least one model segmentation block among the second number of model segmentation blocks.

[0276] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0277] Fig.17 3 is a block diagram of an apparatus 300 for model transmission according to an exemplary embodiment. For example, the apparatus 300 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0278] Reference Fig.17 , the device 300 may include one or more of the following components: a processing component 302 , a memory 304 , a power component 306 , a multimedia component 308 , an audio component 310 , an input / output (I / O) interface 312 , a sensor component 314 , and a communication component 316 .

[0279] The processing component 302 generally controls the overall operation of the device 300, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 302 may include one or more processors 320 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 302 may include one or more modules to facilitate the interaction between the processing component 302 and other components. For example, the processing component 302 may include a multimedia module to facilitate the interaction between the multimedia component 308 and the processing component 302.

[0280] The memory 304 is configured to store various types of data to support operations on the device 300. Examples of such data include instructions for any application or method operating on the device 300, contact data, phone book data, messages, pictures, videos, etc. The memory 304 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0281] The power component 306 provides power to the various components of the device 300. The power component 306 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 300.

[0282] The multimedia component 308 includes a screen that provides an output interface between the device 300 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 308 includes a front camera and / or a rear camera. When the device 300 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0283] The audio component 310 is configured to output and / or input audio signals. For example, the audio component 310 includes a microphone (MIC), and when the device 300 is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 304 or sent via the communication component 316. In some embodiments, the audio component 310 also includes a speaker for outputting audio signals.

[0284] I / O interface 312 provides an interface between processing component 302 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.

[0285] The sensor assembly 314 includes one or more sensors for providing various aspects of the status assessment of the device 300. For example, the sensor assembly 314 can detect the open / closed state of the device 300, the relative positioning of components, such as the display and keypad of the device 300, the sensor assembly 314 can also detect the position change of the device 300 or a component of the device 300, the presence or absence of user contact with the device 300, the orientation or acceleration / deceleration of the device 300, and the temperature change of the device 300. The sensor assembly 314 can include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 314 can also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 314 can also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0286] The communication component 316 is configured to facilitate wired or wireless communication between the device 300 and other devices. The device 300 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 316 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 316 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0287] In an exemplary embodiment, the apparatus 300 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0288] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 304 including instructions, and the instructions can be executed by the processor 320 of the device 300 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0289] Fig.18 4 is a block diagram of a device 400 for model transmission according to an exemplary embodiment. For example, the device 400 can be provided as a server. Fig.18, the apparatus 400 includes a processing component 422, which further includes one or more processors, and a memory resource represented by a memory 432 for storing instructions, such as an application, that can be executed by the processing component 422. The application stored in the memory 432 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 422 is configured to execute the instructions to perform the above method.

[0290] The device 400 may also include a power supply component 426 configured to perform power management of the device 400, a wired or wireless network interface 450 configured to connect the device 400 to a network, and an input / output (I / O) interface 458. The device 400 may operate based on an operating system stored in the memory 432, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or the like.

[0291] It is further understood that in the present disclosure, "plurality" refers to two or more than two, and other quantifiers are similar thereto. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. The singular forms "a", "the" and "the" are also intended to include plural forms, unless the context clearly indicates other meanings.

[0292] It is further understood that the terms "first", "second", etc. are used to describe various information, but such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other, and do not indicate a specific order or degree of importance. In fact, the expressions "first", "second", etc. can be used interchangeably. For example, without departing from the scope of the present disclosure, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information.

[0293] It is further understood that, although the operations are described in a specific order in the drawings in the embodiments of the present disclosure, it should not be understood as requiring the operations to be performed in the specific order shown or in a serial order, or requiring the execution of all the operations shown to obtain the desired results. In certain environments, multitasking and parallel processing may be advantageous.

[0294] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0295] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A model transmission method, characterized in that: Applied to a network device, the method comprises: In response to receiving at least one model distribution request, determining a first model, and segmenting the first model to obtain a first number of model segmentation blocks; Wherein, each model distribution request in the at least one model distribution request corresponds to a model segmentation block.

2. The model transmission method according to claim 1, characterized in that: The model distribution request includes the first model accuracy, the first model reasoning complexity and the first model data volume; In response to receiving at least one model request, determining a first model includes: Determine a second model accuracy, a second model reasoning complexity, and a second model data volume for each of the plurality of models; Based on a predefined operation rule, determine a first operation value of the first model accuracy and the second model accuracy, a second operation value of the first model reasoning complexity and the second model reasoning complexity, and a third operation value of the first model data volume and the second model data volume; The model corresponding to the minimum value of the sum of the first operation value, the second operation value and the third operation value is determined as the first model.

3. The model transmission method according to claim 2, characterized in that: The predefined rules are determined in the following manner: In response to the second model accuracy being greater than or equal to the first model accuracy, determining that the first operation value is zero as the operation rule; or In response to the second model accuracy being less than the first model accuracy, determining the first operation value to be the square of the difference between the first model accuracy and the second model accuracy as the operation rule; or In response to the first model reasoning complexity being greater than or equal to the second model reasoning complexity, determining that the second operation value is zero as the operation rule; or In response to the first model reasoning complexity being less than the second model reasoning complexity, determining the second operation value as the square of the difference between the first model reasoning complexity and the second model reasoning complexity as the operation rule; or In response to the first model data amount being greater than or equal to the second model data amount, determining that the third operation value is zero as the operation rule; or In response to the first model data amount being smaller than the second model data amount, the third operation value is determined as the square of the difference between the first model data amount and the second model data amount as the operation rule.

4. The model transmission method according to claim 1 or 2, characterized in that: The model distribution request includes computing resources and communication resources, and the method further includes: Determine a first model data volume of the first model, and determine a plurality of model segmentation point options based on the first model data volume, computing resources, and communication resources, wherein the model segmentation point options are used to segment the first model.

5. The model transmission method according to claim 4, characterized in that: The method further comprises: A first indication message is sent, wherein the first indication message includes the plurality of model segmentation point options.

6. The model transmission method according to claim 1, characterized in that: The model distribution request includes a first model split point option; The segmenting of the first model comprises: At least one of said first model split point options is determined.

7. The model transmission method according to claim 2, characterized in that: The method further comprises: Determine a merging threshold of the model segments, and determine a difference value between every two of the model segments in the first number of model segments; In response to a first difference value among the difference values ​​being less than or equal to the merging threshold, it is determined that two model segments corresponding to the first difference value are the same, and the model segments are merged to obtain a second number of model segments.

8. The model transmission method according to claim 7, characterized in that: The step of determining a merge threshold of the model segmentation blocks includes: Determining a first model inference complexity and determining a channel quality; The merging threshold is determined to be proportional to the inverse of the product of the first model inference complexity and the channel quality.

9. The model transmission method according to claim 7, characterized in that: The determining of the difference value between every two of the model segmentation blocks in the first number of model segmentation blocks comprises: A matrix of a first number of model partitions is determined, and a difference value between every two of the model partitions is determined based on the matrix.

10. The model transmission method according to claim 7, characterized in that: The method further comprises: A second indication message is sent, wherein the second indication message includes at least one model segmentation block among the second number of model segmentation blocks.

11. The model transmission method according to claim 10, characterized in that: The sending of the second indication message comprises: In response to a first model segmentation block in the second number of model segmentation blocks corresponding to a model distribution request, determining to send a second indication message based on unicast; or In response to a first model segmentation block in the second number of model segmentation blocks corresponding to a plurality of model distribution requests, determining to send a second indication message based on multicast; or In response to a first model segmentation block in the second number of model segmentation blocks corresponding to all model distribution requests, it is determined to send a second indication message based on broadcasting.

12. A model transmission method, characterized in that: Applied to a terminal, the method comprises: Send a model distribution request; wherein the model distribution request corresponds to a model segmentation block.

13. The model transmission method according to claim 12, characterized in that: The method further comprises: receiving a first indication message, wherein the first indication message includes a plurality of model segmentation point options, wherein the model segmentation point options are used to segment the model; A first model segmentation point option is determined based on the first indication message.

14. The model transmission method according to claim 13, characterized in that: The determining a first model segmentation point option based on the first indication message includes: Determining a third number of model segmentation point options from the plurality of model segmentation point options according to the needs of the terminal itself; Determine a first model reasoning complexity and a first model data volume, and determine a third model reasoning complexity and a third model data volume of a model segmentation block corresponding to each of the model segmentation point options in a third number of model segmentation point options; Based on a predefined operation rule, determine fourth operation values ​​of the first model reasoning complexity and the third model reasoning complexity, and fifth operation values ​​of the first model data volume and the third model data volume; The model segmentation point option corresponding to the minimum sum of the fourth operation value and the fifth operation value is determined as the first model segmentation point option.

15. The model transmission method according to claim 14, characterized in that: The predefined operation rules include: In response to the first model reasoning complexity being greater than or equal to the third model reasoning complexity, determining that the fourth operation value is zero as the operation rule; or In response to the first model reasoning complexity being less than the third model reasoning complexity, determining the fourth operation value to be the square of the difference between the first model reasoning complexity and the third model reasoning complexity as the operation rule; or In response to the first model data amount being greater than or equal to the third model data amount, determining that the fifth operation value is zero as the operation rule; or In response to the first model data amount being smaller than the third model data amount, determining the fifth operation value to be the square of the difference between the first model data amount and the third model data amount as the operation rule.

16. The model transmission method according to claim 12, characterized in that: The method further comprises: A second indication message is received, wherein the second indication message includes at least one model segmentation block of the second number of model segmentation blocks.

17. A model transmission device, characterized in that: Applied to network equipment, the device comprises: a determination module, configured to determine a first model in response to receiving at least one model distribution request, and segment the first model to obtain a first number of model segmentation blocks; Wherein, each model distribution request in the at least one model distribution request corresponds to a model segmentation block.

18. A model transmission device, characterized in that: Applied to a terminal, the device comprises: The sending module is used to send a model distribution request; wherein the model distribution request corresponds to a model segmentation block.

19. A model transmission device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to: execute the model transmission method described in any one of claims 1-11, or execute the model transmission method described in any one of claims 12-16.

20. A non-transitory computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by the processor of the mobile terminal, the mobile terminal is enabled to execute the model transmission method described in any one of claims 1-11, or the mobile terminal is enabled to execute the model transmission method described in any one of claims 12-16.

Images