Model transmission method, communication device and system, storage medium, and program product

Abstract

The present disclosure relates to a model transmission method, a communication device and system, a storage medium, and a program product. The model transmission method comprises: sending first information to a second device, wherein the first information is used for indicating a first parameter where a model parameter of an artificial intelligence (AI) model is changed. By means of the embodiments of the present disclosure, the amount of data for model transmission can be reduced.

Description

Model transmission methods, communication equipment, systems, storage media, and program products Technical Field

[0001] This disclosure relates to the field of communication technology, and in particular to model transmission methods, communication devices, systems, storage media, and program products. Background Technology

[0002] With the rapid development of Artificial Intelligence (AI) technology, its applications are becoming increasingly widespread. In communication systems, AI model training can be performed on the terminal side, while inference can be performed on the network side, or vice versa, with AI model data transmitted between the terminal and network sides. Because AI models involve large amounts of data, transmitting the updated AI model using the method of transmitting the complete model when updates occur consumes significant wireless resources. Summary of the Invention

[0003] How to reduce the amount of data transmitted by the model is a problem that needs to be solved.

[0004] This disclosure presents a model transmission method, communication device, system, storage medium, and program product.

[0005] According to a first aspect of the present disclosure, a model transmission method is proposed, executed by a first device, the method comprising: sending first information to a second device, the first information being used to indicate a first parameter of a model parameter of an artificial intelligence (AI) model that has changed.

[0006] According to a second aspect of the present disclosure, a model transmission method is proposed, executed by a second device, the method comprising: receiving first information sent by a first device, the first information being used to indicate a first number of parameter changes in the model parameters of an artificial intelligence (AI) model.

[0007] According to a third aspect of the present disclosure, a communication device is provided for performing the model transmission method of any of the above aspects.

[0008] According to a fourth aspect of the present disclosure, a communication system is proposed, including a first device and a second device, wherein the first device is configured to implement the model transmission method of the first aspect, and the second device is configured to implement the model transmission method of the second aspect.

[0009] According to a fifth aspect of the present disclosure, a storage medium is provided that stores instructions which, when executed on a communication device, cause the communication device to perform the method of the first aspect or the second aspect.

[0010] According to a sixth aspect of the present disclosure, a program product is proposed, including at least one of a program and instructions, wherein when the program or instructions are executed by a communication device, the model transmission method of the first aspect or the second aspect is implemented.

[0011] In this embodiment of the disclosure, the first device sends first information to the second device. The first information is used to indicate the model parameters of the AI ​​model that have changed. The first device transmits the changed parameters of the AI ​​model to the second device, which can reduce the amount of data transmitted by the model, save transmission resources, and reduce the model transmission latency. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings required for the description of the embodiments are introduced below. The following drawings are only some embodiments of this disclosure and do not impose specific limitations on the protection scope of this disclosure.

[0013] Figure 1A is a schematic diagram of the architecture of a communication system according to an embodiment of the present disclosure.

[0014] Figure 1B is a schematic diagram of the LeNet-5 model structure according to an embodiment of the present disclosure.

[0015] Figure 2A is an interactive schematic diagram of a model transmission method according to an embodiment of the present disclosure.

[0016] Figure 2B is an interactive schematic diagram of a model transmission method according to an embodiment of the present disclosure.

[0017] Figure 2C is an interactive schematic diagram of a model transfer method according to an embodiment of the present disclosure.

[0018] Figure 2D is a schematic diagram of a model according to an embodiment of the present disclosure.

[0019] Figure 3 is a schematic flowchart illustrating a model transfer method according to an embodiment of the present disclosure.

[0020] Figure 4 is a schematic flowchart illustrating a model transfer method according to an embodiment of the present disclosure.

[0021] Figure 5A is a schematic diagram of the structure of the first device proposed in an embodiment of this disclosure.

[0022] Figure 5B is a schematic diagram of the structure of the second device proposed in an embodiment of this disclosure.

[0023] Figure 6A is a schematic diagram of the structure of the communication device proposed in an embodiment of this disclosure.

[0024] Figure 6B is a schematic diagram of the chip structure proposed in an embodiment of this disclosure. Detailed Implementation

[0025] This disclosure presents a model transmission method, communication device, system, storage medium, and program product.

[0026] In a first aspect, embodiments of this disclosure propose a model transmission method, executed by a first device, the method comprising: sending first information to a second device, the first information being used to indicate a first parameter of a changed model parameter of an artificial intelligence (AI) model.

[0027] In the above embodiments, the first device sends first information to the second device. The first information is used to indicate the model parameters of the AI ​​model that have changed. The transmission of the changed AI model parameters from the first device to the second device can reduce the amount of data transmitted by the model, save transmission resources, and reduce the model transmission latency.

[0028] In conjunction with some embodiments of the first aspect, in some embodiments, the AI ​​model is based on a bitstream or bytestream representation; the first information includes at least one of the following: position information of the change in the bitstream or the bytestream; the number of bits that have changed in the bitstream; the number of bytes that have changed in the bytestream; and the changed value.

[0029] In some embodiments, in conjunction with the first aspect, the method further includes: compressing the second information to obtain the first information, wherein the second information is obtained by performing an XOR operation on the model parameters before the AI ​​model change and the model parameters after the change.

[0030] In conjunction with some embodiments of the first aspect, in some embodiments, the content of the AI ​​model includes model structure data and model parameters. Without changing the model structure data, the second information is compressed to obtain the first information.

[0031] In conjunction with some embodiments of the first aspect, in some embodiments, the first information includes at least one of the following: the parameter index of the first parameter; the parameter value of the first parameter.

[0032] In conjunction with some embodiments of the first aspect, in some embodiments, the parameter value is the value after the first parameter has been changed, or the parameter value is the change value of the first parameter.

[0033] In conjunction with some embodiments of the first aspect, in some embodiments, the parameter values ​​included in the first information are greater than a threshold value.

[0034] In some embodiments, in conjunction with some embodiments of the first aspect, the method further includes: sending third information to the second device, the third information being used to indicate the use of incremental transmission mode.

[0035] In conjunction with some embodiments of the first aspect, in some embodiments, the third information includes the current version number and the base version number corresponding to the AI ​​model.

[0036] In conjunction with some embodiments of the first aspect, in some embodiments, the first information is used by the second device to update the AI ​​model.

[0037] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes: receiving a first message sent by the second device, the first message indicating that the model update has failed, the first message including at least one of the following: the identifier of the AI ​​model, the current version number, and the baseline version number.

[0038] In conjunction with some embodiments of the first aspect, in some embodiments, when the first device is a terminal device and the second device is a network device, the method further includes at least one of the following: sending fourth information to the network device, the fourth information being used to instruct the terminal device to support uplink transmission using incremental transmission mode; receiving fifth information sent by the network device, the fifth information being used to configure the terminal to perform uplink transmission using incremental transmission mode.

[0039] In conjunction with some embodiments of the first aspect, in some embodiments, when the first device is a network device and the second device is a terminal device, the method further includes at least one of the following: receiving sixth information sent by the terminal device, the sixth information being used to indicate that the terminal device supports downlink reception using incremental transmission mode; and sending seventh information to the terminal device, the seventh information being used to configure the terminal to perform downlink reception using incremental transmission mode.

[0040] Secondly, this disclosure provides a model transmission method, executed by a second device, the method comprising: receiving first information sent by a first device, the first information being used to indicate a first parameter in which the model parameters of an artificial intelligence (AI) model have changed.

[0041] In conjunction with some embodiments of the second aspect, in some embodiments, the AI ​​model is based on a bitstream or bytestream representation; the first information includes at least one of the following: the position information of the change in the bitstream or the bytestream; the number of bits that have changed in the bitstream; the number of bytes that have changed in the bytestream; and the changed value.

[0042] In some embodiments of the second aspect, the method further includes: compressing the second information to obtain the first information, wherein the second information is obtained by performing an XOR operation on the model parameters before and after the AI ​​model change.

[0043] In conjunction with some embodiments of the second aspect, in some embodiments, the content of the AI ​​model includes model structure data and model parameters. Without changing the model structure data, the second information is compressed to obtain the first information.

[0044] In conjunction with some embodiments of the second aspect, in some embodiments, the first information includes at least one of the following: the parameter index of the first parameter; the parameter value of the first parameter.

[0045] In conjunction with some embodiments of the second aspect, in some embodiments, the parameter value is the value after the first parameter has been changed, or the parameter value is the change value of the first parameter.

[0046] In conjunction with some embodiments of the second aspect, in some embodiments, the parameter values ​​included in the first information are greater than a threshold value.

[0047] In conjunction with some embodiments of the second aspect, in some embodiments, the method further includes: receiving third information sent by the first device, the third information being used to indicate the use of an incremental transmission mode.

[0048] In conjunction with some embodiments of the second aspect, in some embodiments, the third information includes the current version number and the base version number corresponding to the AI ​​model.

[0049] In conjunction with some embodiments of the second aspect, in some embodiments, the first information is used by the second device to update the AI ​​model.

[0050] In conjunction with some embodiments of the second aspect, in some embodiments, the method further includes: sending a first message to the first device, the first message indicating that the model update has failed, the first message including at least one of the following: the identifier of the AI ​​model, the current version number, and the base version number.

[0051] In conjunction with some embodiments of the second aspect, in some embodiments, when the first device is a terminal device and the second device is a network device, the method further includes at least one of the following: receiving fourth information sent by the terminal device, the fourth information being used to indicate that the terminal device supports uplink transmission using incremental transmission mode; and sending fifth information to the terminal device, the fifth information being used to configure the terminal to perform uplink transmission using incremental transmission mode.

[0052] In conjunction with some embodiments of the second aspect, in some embodiments, when the first device is a network device and the second device is a terminal device, the method further includes at least one of the following: sending sixth information to the network device, the sixth information being used to instruct the terminal device to support downlink reception using incremental transmission mode; receiving seventh information sent by the network device, the seventh information being used to configure the terminal to perform downlink reception using incremental transmission mode.

[0053] Thirdly, embodiments of this disclosure provide a communication device for performing the model transmission method of the first or second aspect.

[0054] Fourthly, embodiments of this disclosure propose a communication system including a first device and a second device, wherein the first device is configured to implement the model transmission method of the first aspect, and the second device is configured to implement the model transmission method of the second aspect.

[0055] Fifthly, embodiments of this disclosure provide a storage medium storing instructions that, when executed on a communication device, cause the communication device to perform the method of the first aspect or the second aspect.

[0056] In a sixth aspect, embodiments of this disclosure provide a program product, including at least one of a program and instructions, wherein when the program or instructions are executed by a communication device, the communication device performs the model transmission method of the first aspect or the second aspect.

[0057] In a seventh aspect, embodiments of this disclosure provide a chip or chip system. The chip or chip system includes processing circuitry configured to perform the methods described in optional implementations of the first or second aspect.

[0058] It is understood that the aforementioned communication devices, communication systems, storage media, program products, computer programs, chips, or chip systems are all used to execute the methods proposed in the embodiments of this disclosure. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0059] In some embodiments, the terms model transmission method, communication method, information sending method, information reporting method, and information receiving method can be used interchangeably.

[0060] This disclosure is not exhaustive, but merely illustrative of some embodiments, and is not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment can be arbitrarily interchanged. Furthermore, the optional implementation methods in a particular embodiment can be arbitrarily combined; moreover, the embodiments can be arbitrarily combined, for example, some or all steps of different embodiments can be arbitrarily combined, and a particular embodiment can be arbitrarily combined with the optional implementation methods of other embodiments.

[0061] In each of the disclosed embodiments, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of the embodiments are consistent and can be referenced by each other. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0062] The terminology used in the embodiments of this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure.

[0063] In this embodiment of the disclosure, unless otherwise stated, elements expressed in the singular form, such as "a," "an," "the," "the," "the," "the," "the," "the," "this," etc., can mean "one and only one," or "one or more," "at least one," etc. For example, when using articles such as "a," "an," "the," etc. in translation, the noun following the article can be understood as either a singular expression or a plural expression.

[0064] In the embodiments disclosed herein, "multiple" refers to two or more.

[0065] In some embodiments, the terms “at least one of”, “one or more”, “a plurality of”, “multiple”, etc., may be used interchangeably.

[0066] In some embodiments, the notation "at least one of A and B", "A and / or B", "A in one case, B in another", "in response to one case A, in response to another case B", etc., may include the following technical solutions depending on the situation: in some embodiments, A (execute A regardless of B); in some embodiments, B (execute B regardless of A); in some embodiments, execution is selected from A and B (A and B are selectively executed); in some embodiments, A and B (both A and B are executed). The same applies when there are more branches such as A, B, C, etc.

[0067] In some embodiments, the notation "A or B" may include the following technical solutions, depending on the situation: in some embodiments, A (execution of A regardless of B); in some embodiments, B (execution of B regardless of A); in some embodiments, execution is selected from A and B (A and B are selectively executed). The same applies when there are more branches such as A, B, C, etc.

[0068] The prefixes "first," "second," etc., used in the embodiments of this disclosure are merely for distinguishing different descriptive objects and do not impose restrictions on the position, order, priority, quantity, or content of the descriptive objects. The description of the descriptive objects is found in the claims or the context of the embodiments, and the use of prefixes should not constitute unnecessary restrictions. For example, if the descriptive object is a "field," the ordinal numbers preceding "field" in "first field" and "second field" do not restrict the position or order of the "fields." "First" and "second" do not restrict whether the "fields" they modify are in the same message, nor do they restrict the order of "first field" and "second field." Similarly, if the descriptive object is a "level," the ordinal numbers preceding "level" in "first level" and "second level" do not restrict the priority between "levels." Furthermore, the number of descriptive objects is not limited by ordinal numbers and can be one or more. For example, in "first device," the number of "devices" can be one or more. Furthermore, the objects modified by different prefixes can be the same or different. For example, if the object being described is "device", then "first device" and "second device" can be the same device or different devices, and their types can be the same or different. Similarly, if the object being described is "information", then "first information" and "second information" can be the same information or different information, and their content can be the same or different.

[0069] In some embodiments, “including A,” “containing A,” “for indicating A,” and “carrying A” can be interpreted as directly carrying A or indirectly indicating A.

[0070] In some embodiments, the terms “in response to…”, “in response to determining…”, “in the case of…”, “when…”, “if…”, “if…”, etc., can be used interchangeably.

[0071] In some embodiments, the terms “greater than,” “greater than or equal to,” “not less than,” “more than,” “more than or equal to,” “not less than,” “higher than,” “higher than or equal to,” “not lower than,” and “above” can be used interchangeably, as can the terms “less than,” “less than or equal to,” “not greater than,” “less than,” “less than or equal to,” “not more than,” “lower than,” “lower than or equal to,” “not higher than,” and “below”.

[0072] In some embodiments, devices, etc., can be interpreted as physical or virtual, and their names are not limited to the names recorded in the embodiments. Terms such as “device”, “equipment”, “circuit”, “network element”, “node”, “function”, “unit”, “section”, “system”, “network”, “chip”, “chip system”, “entity”, and “subject” can be used interchangeably.

[0073] In some embodiments, "network" can be interpreted as devices included in a network (e.g., access network devices, core network devices, etc.).

[0074] In some embodiments, the terms "access network device (AN device)," "radio access network device (RAN device)," "base station (BS)," "radio base station," "fixed station," "node," "access point," "transmission point (TP)," "reception point (RP)," "transmission / reception point (TRP)," "panel," "antenna panel," "antenna array," "cell," "macro cell," "small cell," "femto cell," "pico cell," "sector," "cell group," "serving cell," "carrier," "component carrier," and "bandwidth part (BWP)" can be used interchangeably.

[0075] In some embodiments, the terms "terminal", "terminal device", "user equipment (UE)", "user terminal", "mobile station (MS)", "mobile terminal (MT)", "subscriber station", "mobile unit", "subscriber unit", "wireless unit", "remote unit", "mobile device", "wireless device", "wireless communication device", "remote device", "mobile subscriber station", "access terminal", "mobile terminal", "wireless terminal", "remote terminal", "handset", "user agent", "mobile client", and "client" can be used interchangeably.

[0076] In some embodiments, access network devices, core network devices, or network devices can be replaced by terminals. For example, embodiments of this disclosure can also be applied to structures where communication between access network devices, core network devices, or network devices and terminals is replaced by communication between multiple terminals (e.g., device-to-device (D2D), vehicle-to-everything (V2X), etc.). In this case, the structure can also be configured such that the terminal has all or part of the functions of the access network device. Furthermore, terms such as "uplink" and "downlink" can be replaced with terms corresponding to communication between terminals (e.g., "sidelink"). For example, uplink channel, downlink channel, etc., can be replaced with sidelink channel, and uplink link, downlink, etc., can be replaced with sidelink link.

[0077] In some embodiments, the terminal may be replaced by an access network device, a core network device, or a network device. In this case, the access network device, core network device, or network device may also be configured to have all or some of the functions of the terminal.

[0078] In some embodiments, the acquisition of data, information, etc., may comply with the laws and regulations of the country where the location is situated.

[0079] In some embodiments, data, information, etc., may be obtained with the user's consent.

[0080] Furthermore, each element, each row, or each column in the table of this disclosure can be implemented as an independent embodiment, and any combination of any element, any row, or any column can also be implemented as an independent embodiment.

[0081] Figure 1A is a schematic diagram of the architecture of a communication system according to an embodiment of the present disclosure.

[0082] As shown in Figure 1A, the communication system 100 includes a first device 101 and a second device 102.

[0083] In some embodiments, the first device 101 is a device that sends an AI model, and the second device 102 is a device that receives an AI model.

[0084] In some embodiments, the first device 101 is a terminal device; the second device 102 is a network device, such as an access network device.

[0085] In other embodiments, the first device 101 is a network device, such as an access network device; and the second device 102 is a terminal device.

[0086] In some embodiments, the terminal device may be, for example, a user equipment (UE), including, at least one of, but not limited to, a mobile phone, a wearable device, an Internet of Things device, a car with communication capabilities, a smart car, a tablet computer, a computer with wireless transceiver capabilities, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal device in industrial control, a wireless terminal device in self-driving, a wireless terminal device in remote medical surgery, a wireless terminal device in a smart grid, a wireless terminal device in transportation safety, a wireless terminal device in a smart city, and a wireless terminal device in a smart home.

[0087] In some embodiments, a network device can be a functional network element within a core network device. The core network device can be a single device, including a first network element, a second network element, etc., or it can be multiple devices or a group of devices, each including all or part of the first network element, the second network element, etc. Network elements can be virtual or physical. The core network includes, for example, at least one of an Evolved Packet Core (EPC), a 5G Core Network (5GCN), and a Next Generation Core (NGC).

[0088] In some embodiments, the network device may include at least one of an access network device and a core network device.

[0089] In some embodiments, the access network device is, for example, a node or device that connects a terminal to a wireless network. The access network device may include, but is not limited to, at least one of the following in a 5G communication system: evolved Node B (eNB), next-generation eNB (ng-eNB), next-generation Node B (gNB), node B (NB), home node B (HNB), home evolved node B (HeNB), radio backhaul device, radio network controller (RNC), base station controller (BSC), base transceiver station (BTS), base band unit (BBU), mobile switching center, base station in a 6G communication system, open RAN, cloud RAN, base station in other communication systems, and access node in a Wi-Fi system.

[0090] In some embodiments, the technical solutions of this disclosure can be applied to the Open RAN architecture. In this case, the interfaces between or within access network devices involved in the embodiments of this disclosure can be transformed into internal interfaces of Open RAN. The processes and information interactions between these internal interfaces can be implemented by software or programs.

[0091] In some embodiments, the access network device may be composed of a central unit (CU) and a distributed unit (DU). The CU may also be called a control unit. The CU-DU structure can separate the protocol layer of the access network device. Some of the protocol layer functions are centrally controlled by the CU, while the remaining part or all of the protocol layer functions are distributed in the DU and centrally controlled by the CU. However, this is not the only possibility.

[0092] In some embodiments, a core network device may be a single device comprising one or more network elements, or it may be multiple devices or a group of devices, each comprising all or part of the aforementioned one or more network elements. Network elements may be virtual or physical. The core network may include, for example, at least one of an Evolved Packet Core (EPC), a 5G Core Network (5GCN), or a Next Generation Core (NGC).

[0093] It is understood that the communication system described in this disclosure is for the purpose of more clearly illustrating the technical solutions of this disclosure, and does not constitute a limitation on the technical solutions proposed in this disclosure. As those skilled in the art will know, with the evolution of system architecture and the emergence of new business scenarios, the technical solutions proposed in this disclosure are also applicable to similar technical problems.

[0094] The following embodiments of this disclosure can be applied to the communication system 100 shown in FIG1A, or to some of the main bodies, but are not limited thereto. The main bodies shown in FIG1A are illustrative. The communication system may include all or some of the main bodies in FIG1A, or it may include other main bodies outside of FIG1A. The number and form of each main body are arbitrary. Each main body may be physical or virtual. The connection relationship between the main bodies is illustrative. The main bodies may not be connected or may be connected. The connection can be in any way, it can be a direct connection or an indirect connection, it can be a wired connection or a wireless connection.

[0095] The embodiments disclosed herein can be applied to Long Term Evolution (LTE), LTE-Advanced (LTE-A), LTE-Beyond (LTE-B), SUPER 3G, IMT-Advanced, 4th generation mobile communication system (4G), 5th generation mobile communication system (5G), 5G new radio (NR), 6th generation mobile communication system (6G), Future Radio Access (FRA), New-Radio Access Technology (RAT), New Radio (NR), New radio access (NX), Future generation radio access (FX), Global System for Mobile communications (GSM), CDMA2000, Ultra Mobile Broadband (UMB), IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), and IEEE 802.20, Ultra-Wideband (UWB), Bluetooth (a registered trademark), Public Land Mobile Network (PLMN) networks, Device-to-Device (D2D) systems, Machine-to-Machine (M2M) systems, Internet of Things (IoT) systems, Vehicle-to-Everything (V2X) systems, systems utilizing other communication methods, and next-generation systems built upon them, etc. Furthermore, multiple systems can be combined (e.g., a combination of LTE or LTE-A with 5G).

[0096] Machine learning algorithms are one of the important methods for implementing artificial intelligence technology. Machine learning can obtain models through large amounts of training data, and these models can then predict events. In many fields, machine learning models can achieve highly accurate predictions. In 5G systems, AI can be used for prediction and inference, improving system performance.

[0097] In some embodiments, when the AI ​​model is trained on the network side and AI inference is run on the terminal side, the network device needs to transmit the AI ​​model to the terminal. When the AI ​​model is trained on an Over-The-Top (OTT) server and AI inference is run on the network side, the terminal needs to transmit the AI ​​model received from the OTT server to the network device.

[0098] In some embodiments, an AI model has a specific model structure and corresponding parameters / weights. The model structure can be based on a specific backbone, such as a Convolutional Neural Network (CNN) or a Transformer. AI models with the same backbone can also have different structures, such as different numbers of layers. There are various CNN-based AI model architectures, such as the LeNet-5 model, whose structure is shown in Figure 1B. Referring to Figure 1B, the LeNet-5 model can include an input layer, C1 layer, S2 layer, C3 layer, S4 layer, C5 layer, F6 layer, and an output layer. The input layer can take a 32×32 image as input. Layer C1 performs convolution on the input image, resulting in six 28×28 feature maps. Layer S2 performs max pooling, resulting in six 14×14 feature maps. Layer C3 performs convolution, resulting in sixteen 10×10 feature maps. Layer S4 performs max pooling, resulting in sixteen 5×5 feature maps. Layer C5 performs convolution, resulting in a one-dimensional vector with 120 parameters. Layer F6 performs classification, resulting in a one-dimensional vector with 84 parameters; layer F6 is a fully connected (FC) layer. The output layer performs classification, resulting in a one-dimensional vector with 10 parameters; the output layer is also a fully connected layer.

[0099] For an AI model, it can be retrained with newly acquired data to update the model, such as fine-tuning it. The updated model also needs to be transmitted between network devices and terminals.

[0100] There are several ways to transmit AI models. One method is to use a standardized format, such as the Open Neural Network Exchange (ONNX). ONNX is a protobuf (Protocol Buffer) file that contains the model graph (i.e., the model structure) and its corresponding weights. Protobuf is a method for structured data serialization. In the ONNX model format, the model structure and model parameters are mixed together.

[0101] There are multiple use cases for AI model transmission. One such use case is that the model is trained and stored on the network side, and the model transmission uses an open format. The terminal already knows the model structure; that is, the network and the terminal have identified the exact model structure before the model transmission, and the terminal has explicitly indicated its support for that model structure.

[0102] Machine learning algorithms often use tensors as their basic data structure. One way to define a tensor is as a multidimensional array. A zero-dimensional tensor is a scalar, a one-dimensional tensor is a vector, and a two-dimensional tensor is a matrix.

[0103] The values ​​in a multidimensional array are called tensor components, and their positions in the array are represented by indices.

[0104] Key properties of tensors include:

[0105] Dimension / order: This refers to the array dimension, or the number of indexes needed to identify each component;

[0106] Shape: The number of elements in each dimension of a tensor. For example, the shape of a 3×5 matrix (a two-dimensional tensor) is (3, 5).

[0107] Data type: The type of data contained in a tensor, such as float16 (16-bit floating-point number), int8 (8-bit integer), int4 (4-bit integer), etc.

[0108] The exclusive OR (XOR) operation between two bits x and y is defined as follows:

[0109] The XOR operation for two integers is defined as the XOR operation for their corresponding bits. For example...

[0110] AI models can be quite large, ranging from several MB to several GB. When updating an AI model, the content transmitted may not have changed completely. Transmitting updated AI models using the method of transmitting the complete model would consume significant wireless resources.

[0111] Figure 2A is an interactive schematic diagram of a model transfer method according to an embodiment of the present disclosure. As shown in Figure 2A, the present disclosure relates to a model transfer method, which includes:

[0112] In step S2101, the first device 101 sends the first information to the second device 102.

[0113] In some embodiments, the second device 102 receives first information sent by the first device 101.

[0114] In some embodiments, the first information is used to indicate the first parameter of the model parameters of the Artificial Intelligence (AI) model that has changed. That is, the first parameter refers to the model parameter in the AI ​​model that has changed, and there may be one or more first parameters.

[0115] In some embodiments, the first device 101 is a device for sending model data, and the second device 102 is a device for receiving model data; the AI ​​model is trained on the first device 101 and inferred (or applied) on the second device 102; the first device 101 sends the AI ​​model data to the second device 102, and the second device 102 deploys the AI ​​model; when the AI ​​model is updated, the first device 101 can send the changed model parameters to the second device 102, and the second device 102 updates the AI ​​model according to the information sent by the first device 101.

[0116] In some embodiments, the first device 101 is a terminal device; the second device 102 is a network device, such as an access network device. In this case, the AI ​​model is trained on the terminal device and inferred on the network device. The terminal device sends the AI ​​model data to the network device. When the AI ​​model is updated, the terminal device sends the changed model parameters to the network device, and the network device updates the AI ​​model according to the information sent by the terminal device.

[0117] In other embodiments, the first device 101 is a network device, such as an access network device; the second device 102 is a terminal device; in this case, the AI ​​model is trained on the network device and inferred on the terminal device, and the network device sends the content of the AI ​​model to the terminal device; when the AI ​​model is updated, the network device sends the changed model parameters to the terminal device, and the terminal device updates the AI ​​model according to the information sent by the network device.

[0118] In some embodiments, when the AI ​​model is updated, the model parameters that change in the AI ​​model can be the updated model parameters of the updated AI model relative to the baseline AI model, wherein the baseline AI model can be the AI ​​model before the update or a specified version of the AI ​​model.

[0119] In some embodiments, the AI ​​model is represented based on a bit stream or a byte stream.

[0120] In some embodiments, the first information includes at least one of the following: position information of the change in the bit stream or byte stream; the number of bits that changed in the bit stream; the number of bytes that changed in the byte stream; and the changed value.

[0121] In some embodiments, the first information indicates the location of the change and the changed value, which facilitates the second device in updating the AI ​​model.

[0122] In some embodiments, the first information may be a list, where each element in the list corresponds to a changed model parameter.

[0123] In some embodiments, the AI ​​model is based on a bitstream representation. The first information includes at least one of the following: the location information of the change in the bitstream, the number of bits that have changed in the bitstream, and the changed value of each changed bit. The first device sends the first information to the second device, and the second device updates the value at the corresponding location based on the location information of the change, the number of changes, and the changed value, thereby updating the AI ​​model.

[0124] In some embodiments, the AI ​​model is based on a byte stream representation. The first information includes at least one of the following: the location information of the change in the byte stream, the number of bytes in the byte stream that have changed, and the changed value of each changed byte. The first device sends the first information to the second device, and the second device updates the value at the corresponding location based on the location information of the change, the number of changes, and the changed value, thereby updating the AI ​​model.

[0125] The following explanation uses an AI model based on byte stream representation as an example, but this disclosure does not limit the representation form of the AI ​​model.

[0126] Figure 2D is a schematic diagram of a model according to an embodiment of the present disclosure. Taking the model structure shown in Figure 2D as an example, the model has an input of X and an output of Y. Both input X and output Y are vectors with two elements. The operation performed by the model is Y = Add(MatMul(A,X),C), that is, Y = AX + C, where MatMul(A,X) is the matrix multiplication of A and X, and Add(AX,C) is the sum of AX and C. The model parameters include matrix A and vector C, where matrix A is a 2×2 matrix and vector C is a vector with two elements. In Figure 2D, parameter "B<2×2>" represents matrix A, and parameter "B <2> " represents vector C.

[0127] For example, the data types in the model are all 32-bit floating-point numbers, and matrix A is... Vector C is If the model is stored in ONNX format, the model size is 140 bytes. The model content (for ease of distinction, this model content is referred to as content 1) is as follows:

[0128] The above content is in hexadecimal format. Hexadecimal 0x08 represents the first byte of the model content, 0x09 the second byte, 0x3a the third byte, and so on. The numbers on the left of the above content represent offset values; for example, 00000010 represents the offset value of this line of data. The numbers on the right of the above content represent the ASCII codes corresponding to the model content, where "." represents a non-printable character; for example, hexadecimal 0x08 corresponds to the non-printable character ".", and ":" is a printable character; for example, hexadecimal 0x3a corresponds to the printable character ":".

[0129] For example, the AI ​​model mentioned above has been updated, while the model structure remains the same, but the model parameters have changed, and matrix A has been changed to... Vector C changed to The bolded numbers indicate the changed parameters; for example, parameter 0.4 in matrix A is changed to 0.6, and parameter -0.1 in vector C is changed to 0.2. Therefore, the updated AI model size remains 140 bytes. The updated AI model content (referred to as content 2 for clarity) is shown below, with changes relative to content 1 indicated by underlined lines.

[0130] Taking the aforementioned changed model parameters as an example, the first information sent by the first device to the second device may include the position of the changed model parameters in the byte stream, the number of bytes, and the changed values. For example, the first information is a list containing two elements. One element indicates (0x43, 0x04, [0x9a, 0x99, 0x19, 0x3f]), where 0x43 represents the starting position of the change, 0x04 represents the number of bytes changed, and [0x9a, 0x99, 0x19, 0x3f] represents the changed values ​​corresponding to these four positions; that is, the four bytes of data starting from position 0x43 (where the position of the first byte of the AI ​​model is 0) are updated to [0x9a, 0x99, 0x19, 0x3f]; the second element indicates (0x5c, 0x02, [0x4c, 0x3e]), that is, the two bytes of data starting from position 0x5c are updated to [0x4c, 0x3e]. As this example shows, the first piece of information requires 10 bytes (assuming that the location information and the number of changed bits are each indicated by one byte), which greatly reduces the amount of data transmitted compared to using 140 bytes.

[0131] It is understandable that the above example uses hexadecimal as an example. In practical applications, those skilled in the art can also use decimal or other number systems.

[0132] It is understood that the above-mentioned model structure, model content, and changed model parameters are all illustrative examples, and this disclosure does not limit the model structure, model content, or changed model parameters.

[0133] In this embodiment of the disclosure, the first device sends the changed location information, the number of changes, and the changed values ​​to the second device. Compared with sending all the changed model parameters to the second device, this can reduce the amount of data transmitted and improve transmission efficiency.

[0134] In some embodiments, the first device compresses the second information to obtain the first information, that is, the first information is obtained by compressing the second information, and the second information is obtained by performing an XOR operation on the model parameters before and after the AI ​​model change.

[0135] In some embodiments, when the AI ​​model is updated, the first device can perform an XOR operation on the model parameters before the change and the model parameters after the change to obtain second information, and compress the second information to obtain first information; the first device can send the first information to the second device, the second device can decompress the first information, and perform an XOR operation on the decompressed information and the model parameters before the change to obtain the changed model data parameters.

[0136] In this embodiment of the disclosure, an XOR operation is performed on the model parameters before and after the change. The model parameters can be represented based on bit streams or byte streams. For the parts of the parameters that have not changed, the result of the XOR operation is 0. After a large number of repeated 0s are compressed, the amount of data is significantly reduced, thereby reducing the amount of data transmitted.

[0137] In some embodiments, the model parameters before the change can be the model parameters of the baseline model, such as the above content 1; the model parameters before the change can also be the model parameters of the previous version of the current version.

[0138] In some embodiments, any compression algorithm can be used to compress the second information, such as the DEFLATE algorithm, which is a lossless compression algorithm.

[0139] In some embodiments, the content of the AI ​​model includes model structure data and model parameters. If the model structure data remains unchanged, the first device compresses the second information to obtain the first information.

[0140] In some embodiments, the content of the AI ​​model includes data related to the model structure (hereinafter referred to as model structure data) and data related to the model parameters. Generally, when the AI ​​model is updated, the model structure data does not change. If the model structure data does not change, the first device can perform an XOR operation on the model parameters before and after the change to obtain second information, and then compress the second information to obtain the first information. This reduces the amount of data processing and improves data processing efficiency.

[0141] For example, the model structure and model parameters of an AI model can be distinguished at the content transmission level. For instance, in a model content file, the first part is the model structure and the second part is the model parameters. In this case, only the model parameters before and after the update are XORed and compressed before transmission.

[0142] In some embodiments, the first device may also perform an XOR operation on the content of the AI ​​model before and after the update, and compress the result to obtain the first information. This involves performing an XOR operation on the entire content of the AI ​​model, which is applicable both when model parameters change and when both the model structure data and model parameters change.

[0143] For example, taking content 1 above as the model content before the update (also known as the model content of the baseline model), and taking content 2 above as the model content after the update, the result of XORing content 1 and content 2 is as follows:

[0144] As can be seen, after performing an XOR operation on content 1 and content 2, the result of the XOR operation is 0 for the parts of the data that have not changed. These repeated 0s are compressed in the subsequent process, which can significantly reduce the amount of data.

[0145] In some embodiments, the first information includes at least one of the following: the parameter index of the first parameter; the parameter value of the first parameter.

[0146] In some embodiments, the first information can be a list containing multiple elements, where each element corresponds to a changed parameter. For ease of description, the changed parameter is referred to as the first parameter, and there can be one or more first parameters. Each element in the list includes the parameter index and parameter value of the first parameter, where the parameter index indicates the position of the first parameter in the model data. Unlike the embodiments described above, this embodiment indicates the changed parameter at the level of model content. This parameter can be the smallest data unit in the AI ​​model, such as a 2×2 matrix containing 4 elements, each of which can be considered a parameter.

[0147] In some embodiments, the parameter value is the value after the first parameter is changed, or the parameter value is the change value of the first parameter.

[0148] In some embodiments, there may be one or more parameters that change, that is, there may be one or more first parameters. The parameter value corresponding to each first parameter can be the changed value of the first parameter, that is, the parameter value used by the updated AI model, which is an "absolute" value. For example, if the value of the first parameter changes from 0.4 to 0.6, then the parameter value corresponding to the first parameter is 0.6. The first device sends the index of the first parameter and the changed value to the second device, and the second device updates the model.

[0149] For example, the updated model parameters are: (Bold numbers indicate changed parameters). Using the smallest data unit in the AI ​​model as the parameter, the model contains 6 parameters, with indexes of 0, 1, 2, 3, 4, and 5, corresponding to values ​​of 0.7, -0.9, 0.6, 1.2, 0.8, and 0.2 respectively. The first piece of information can indicate a list containing the following elements: (2, 0.6), (5, 0.2), meaning the parameter at index 2 is updated to 0.6, and the parameter at index 5 is updated to 0.2.

[0150] In other embodiments, there may be one or more parameters that change; that is, there may be one or more first parameters. The parameter value corresponding to each first parameter can be the change value of that first parameter, that is, the change value of the updated parameter value relative to the parameter value before the update (e.g., the parameter value of the baseline model), which is a "relative" value. For example, if the value of the first parameter changes from 0.4 to 0.6, then the parameter value corresponding to the first parameter is 0.2. The first device sends the index and change value of the first parameter to the second device, and the second device updates the model.

[0151] For example, the model parameters before the update were The updated model parameters are: (Bold numbers indicate changed parameters). Using the smallest data unit in the AI ​​model as the parameter, the model contains 6 parameters, with indices 0, 1, 2, 3, 4, and 5. The values ​​before the update were 0.7, -0.9, 0.4, 1.2, 0.8, and -0.1, respectively, and the values ​​after the update are 0.7, -0.9, 0.6, 1.2, 0.8, and 0.2. The first piece of information can indicate a list containing the following elements: (2, 0.2), (5, 0.3), meaning the parameter at index 2 changes by 0.6 - 0.4 = 0.2, and the parameter at index 5 changes by 0.2 – (-0.1) = 0.3.

[0152] In this embodiment of the disclosure, the parameter value of the first parameter included in the first information is the change value of the first parameter. The parameter value can be indicated using a smaller number of bits, thereby further reducing the amount of data transmitted.

[0153] In some embodiments, the parameter value included in the first information is greater than a threshold value.

[0154] In some embodiments, a threshold value can be set. When the parameter value (the changed value or the change in value) is less than or equal to the threshold value, the first information does not include the information corresponding to the parameter value. That is, the first information includes the parameter value only when the parameter value is greater than the threshold value.

[0155] In some embodiments, the threshold value may be configured by the network device or predefined by the protocol.

[0156] In some embodiments, when the absolute value (absolute value in a mathematical sense) of the change in a parameter value is less than or equal to a threshold value, the first information does not include the relevant information (parameter index and parameter value) of the parameter corresponding to that parameter value, further reducing the amount of data transmitted for the AI ​​model.

[0157] For example, with a threshold of 0.2, if the parameter value is the changed value after the first parameter has changed, the parameter at index 2 is updated to 0.6, and the parameter at index 5 is updated to 0.2. Since the parameter at index 2 is updated to 0.6, which is greater than the threshold of 0.2, the first information includes the parameter value of 0.6 corresponding to the parameter at index 2. Since the parameter at index 5 is updated to 0.2, which is not greater than the threshold of 0.2, the first information does not include the parameter value of 0.2 corresponding to the parameter at index 5. That is, the first information only includes the element (2, 0.6).

[0158] For example, with a threshold value of 0.2, considering the change in the first parameter value, the parameter at index 2 has a change value of 0.2, and the parameter at index 5 has a change value of 0.3. Since the change value of the parameter at index 2 is 0.2, which is less than the threshold value of 0.2, the first information does not include the change value of 0.2 corresponding to the parameter at index 2. Since the change value of the parameter at index 5 is 0.3, which is greater than the threshold value of 0.2, the first information includes the change value of 0.3 corresponding to the parameter at index 5. That is, the first information only includes the element (5, 0.3).

[0159] In step S2102, the first device 101 sends third information to the second device 102.

[0160] In some embodiments, the second device 102 receives third information sent by the first device 101.

[0161] In some embodiments, the third information is used to indicate the use of incremental transmission.

[0162] In some embodiments, steps S2101 and S2102 may be performed in an alternate order or simultaneously.

[0163] In other words, step S2101 can be executed first, followed by step S2102. For example, the first device can send the first information to the second device first, and then instruct the second device to use the incremental transmission method. Alternatively, step S2102 can be executed first, followed by step S2101. For example, the first device can instruct the second device to use the incremental transmission method first, and then send the first information to the second device. Alternatively, steps S2101 and S2102 can be executed simultaneously. For example, the third information is included in the first information, and the first device instructs the second device to use the incremental transmission method while sending the first information.

[0164] In some embodiments, incremental transmission refers to transmitting only information indicating changes to the AI ​​model. The first device can determine whether to use incremental transmission. If it determines to use incremental transmission, the first device sends a third message to the second device, instructing it to use incremental transmission. If it determines not to use incremental transmission, the first device may not send the third message, or it may send an instruction indicating not to use incremental transmission. In the case of not using incremental transmission, the first device transmits all parameters of the AI ​​model to the second device.

[0165] In some embodiments, the third information includes the current version number and the baseline version number of the AI ​​model.

[0166] In some embodiments, when the first device determines that incremental transmission mode is to be used, the first device sends third information to the second device. The third information indicates that incremental transmission mode is to be used. The third information may include the current version number and the base version number corresponding to the AI ​​model. The current version number refers to the version number used after the AI ​​model is updated. The base version number may be a predefined version number, or the version number used before the AI ​​model is updated, or the version number used in the most recent transmission, or the version number determined in advance by the first device and the second device.

[0167] In some embodiments, when transmitting AI model data, the first device indicates the version number of the current model to the second device, and optionally indicates the version number of the model on which the current model is based, i.e., the base version number. If the base version number is indicated, it is considered as indicating that an incremental transmission method is used, i.e., the AI ​​model corresponding to the version number of the model on which the model is based is the base model. The base model can be a model that has been agreed upon offline, a related AI model from the most recent transmission, or a related model from the most recent transmission whose transmission has been confirmed by the receiver. If the version number of the model on which the current model is based is not indicated, i.e., the base version number is not indicated, it is considered as indicating that an incremental transmission method is not used.

[0168] In some embodiments, the first information is used by the second device to update the AI ​​model.

[0169] In some embodiments, the second device updates the AI ​​model based on the first information.

[0170] In some embodiments, the first device determines to use an incremental transmission method, the first device sends first information to the second device, and the second device updates the basic model based on the first information.

[0171] In step S2103, the second device 102 sends a first message to the first device 101.

[0172] In some embodiments, the first device 101 receives a first message sent by the second device 102.

[0173] In some embodiments, a first message is used to indicate that the model update has failed, and the first message includes at least one of the following: the identifier of the AI ​​model, the current version number, and the baseline version number.

[0174] In some embodiments, the second device updates the base model based on the first information. If the second device fails to update, the second device reverts the AI ​​model to the base model or the most recently used AI model, and sends a first message indicating that the AI ​​model update failed to occur to the first device.

[0175] In some embodiments, the first message includes an identifier of the AI ​​model, which helps the first device identify the AI ​​model whose update failed; the first message includes the current version number and the base version number, which helps the first device identify the version on which the update failed and the version on which it was based.

[0176] For example, when the first device is a network device and the second device is a terminal device, when the terminal device receives an AI model using incremental transmission, it updates a copy of the base model based on the content of the incremental transmission. If the terminal device fails to update the AI ​​model, it reverts to the base model or the most recently used AI model and reports the update failure message to the network device. The reported message may include the identifier of the AI ​​model, the version number of the base model, and the version number of the updated model.

[0177] The model transmission method provided in this embodiment involves a first device sending first information to a second device. The first information indicates that the model parameters of the AI ​​model have changed. The first device transmits the information about the changes in the AI ​​model to the second device, which can reduce the amount of data transmitted by the model, save transmission resources, and reduce model transmission latency.

[0178] The model transfer method involved in the embodiments of this disclosure may include at least one of steps S2101 to S2103. For example, step S2101 may be implemented as an independent embodiment, step S2102 may be implemented as an independent embodiment, and step S2103 may be implemented as an independent embodiment, but is not limited thereto.

[0179] In some embodiments, steps S2101 and S2102 may be performed in an alternate order or simultaneously.

[0180] In some embodiments, step S2102 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0181] In some embodiments, step S2103 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0182] In some embodiments, other alternative implementations may be described before or after the specification corresponding to FIG2A.

[0183] Figure 2B is an interactive schematic diagram of a model transmission method according to an embodiment of the present disclosure. Figure 2B is illustrated using an example where the first device is a terminal device and the second device is a network device.

[0184] As shown in Figure 2B, this disclosure relates to a model transfer method, which includes:

[0185] Step S2201: The terminal device sends the fourth information to the network device.

[0186] In some embodiments, the network device receives fourth information sent by the terminal device.

[0187] In some embodiments, the fourth information is used to indicate that the terminal device supports uplink transmission using incremental transmission.

[0188] In some embodiments, the terminal device supports uplink transmission using incremental transmission, which means that when the AI ​​model is updated, the terminal device can send information related to the changed model parameters to the network device.

[0189] In some embodiments, the terminal device is the sender of model data and the network device is the receiver of model data. Before the terminal device sends the first information, the terminal device may send the fourth information to the network device to report to the network device that the terminal device supports uplink transmission using incremental transmission mode. The fourth information may be carried in the terminal capability signaling.

[0190] In some embodiments, if the terminal device does not support uplink transmission using incremental transmission, the terminal device may report to the network device that it does not support uplink transmission using incremental transmission, or the terminal device may not report the fourth information to the network device. In this case, the network device determines that the terminal device does not support uplink transmission using incremental transmission.

[0191] In some embodiments, the terminal device does not support uplink transmission using incremental transmission method, which means that when the AI ​​model is updated, the terminal device sends all the updated model data to the network device.

[0192] In some embodiments, step S2201 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0193] In step S2202, the network device sends the fifth information to the terminal device.

[0194] In some embodiments, the terminal device receives fifth information sent by the network device.

[0195] In some embodiments, the fifth information is used to configure the terminal device to use incremental transmission for uplink transmission.

[0196] In some embodiments, if the network device also supports incremental transmission, the network device can send fifth information to the terminal device based on the fourth information to configure the terminal device to use incremental transmission for uplink transmission. If the network device does not support incremental transmission, the network device can send a message to the terminal device based on the fourth information to configure the terminal not to use incremental transmission for uplink transmission.

[0197] In some embodiments, if the network device configures the terminal device to use incremental transmission for uplink transmission, then when the AI ​​model is subsequently updated, the terminal device sends information related to the changed model parameters, i.e., first information, to the network device.

[0198] In some embodiments, step S2202 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0199] Step S2203: The terminal device sends the first information to the network device.

[0200] In some embodiments, the network device receives first information sent by the terminal device.

[0201] The optional implementation of step S2203 can be found in the optional implementation of step S2101 in Figure 2A, and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0202] Step S2204: The terminal device sends third information to the network device.

[0203] In some embodiments, the network device receives third information sent by the terminal device.

[0204] The optional implementation of step S2204 can be found in the optional implementation of step S2102 in Figure 2A, and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0205] Step S2205: The network device sends the first message to the terminal device.

[0206] In some embodiments, the terminal device receives a first message sent by the network device.

[0207] The optional implementation of step S2205 can be found in the optional implementation of step S2103 in Figure 2A, and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0208] The model transfer method involved in the embodiments of this disclosure may include at least one of steps S2201 to S2205. For example, step S2203 may be implemented as an independent embodiment, step S2204 may be implemented as an independent embodiment, and step S2205 may be implemented as an independent embodiment, but is not limited thereto.

[0209] In some embodiments, steps S2203 and S2204 may be performed in an alternate order or simultaneously.

[0210] In some embodiments, steps S2201 and S2202 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0211] In some embodiments, steps S2204 and S2205 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0212] In some embodiments, other optional implementations may be described before or after the specification corresponding to FIG2B.

[0213] Figure 2C is an interactive schematic diagram of a model transmission method according to an embodiment of the present disclosure. Figure 2C is illustrated using an example where the first device is a network device and the second device is a terminal device.

[0214] As shown in Figure 2C, this disclosure relates to a model transfer method, which includes:

[0215] Step S2301: The terminal device sends the sixth information to the network device.

[0216] In some embodiments, the network device receives sixth information sent by the terminal device.

[0217] In some embodiments, the sixth information is used to indicate that the terminal device supports downlink reception using incremental transmission.

[0218] In some embodiments, the terminal device supports downlink reception using incremental transmission, which means that when the AI ​​model is updated, the terminal device can receive information related to the changed model parameters sent by the network device, and the terminal device can update the AI ​​model based on the received information.

[0219] In some embodiments, the network device is the sender of model data and the terminal device is the receiver of model data. Before the network device sends the first information, the terminal device may send a sixth information to the network device to report to the network device that the terminal device supports downlink reception using incremental transmission mode. The sixth information may be carried in the terminal capability signaling.

[0220] In some embodiments, if the terminal device does not support downlink reception using incremental transmission, the terminal device may report to the network device that it does not support downlink reception using incremental transmission, or the terminal device may not report the sixth information to the network device. In this case, the network device determines that the terminal device does not support downlink reception using incremental transmission.

[0221] In some embodiments, the terminal device does not support downlink reception using incremental transmission, which means that when the AI ​​model is updated, the terminal device cannot receive information related to the changed model parameters sent by the network device, or the terminal device cannot update the AI ​​model based on the received information.

[0222] In some embodiments, the sixth information sent by the terminal device to the network device can be sent separately from the fourth information described above. For example, the terminal device may report in its terminal capability signaling that it supports incremental transmission for both uplink transmission and downlink reception. Alternatively, the sixth information sent by the terminal device to the network device can be sent simultaneously with the fourth information described above. For example, the terminal device may jointly report in its terminal capability signaling that it supports incremental transmission for both uplink transmission and downlink reception.

[0223] In some embodiments, step S2301 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0224] In step S2302, the network device sends the seventh information to the terminal device.

[0225] In some embodiments, the terminal device receives seventh information sent by the network device.

[0226] In some embodiments, the seventh information is used to configure the terminal to use incremental transmission for downlink reception.

[0227] In some embodiments, if the network device also supports incremental transmission, the network device can send the seventh information to the terminal device based on the sixth information, configuring the terminal device to use incremental transmission for downlink reception. If the network device does not support incremental transmission, the network device can send the seventh information to the terminal device based on the sixth information, configuring the terminal device not to use incremental transmission for downlink reception.

[0228] In some embodiments, if the network device configures the terminal device to use incremental transmission for downlink reception, then when the AI ​​model is subsequently updated, the network device sends information related to the changed model parameters, i.e., first information, to the terminal device. The terminal device receives the first information and updates the AI ​​model based on the first information.

[0229] In some embodiments, step S2302 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0230] In step S2303, the network device sends the first information to the terminal device.

[0231] In some embodiments, the terminal device receives first information sent by the network device.

[0232] The optional implementation of step S2303 can be found in the optional implementation of step S2101 in Figure 2A, and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0233] Step S2304: The network device sends third information to the terminal device.

[0234] In some embodiments, the terminal device receives third information sent by the network device.

[0235] The optional implementation of step S2304 can be found in the optional implementation of step S2102 in Figure 2A, and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0236] Step S2305: The terminal device sends the first message to the network device.

[0237] In some embodiments, the network device receives a first message sent by the terminal device.

[0238] The optional implementation of step S2305 can be found in the optional implementation of step S2103 in Figure 2A, and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0239] The model transfer method involved in the embodiments of this disclosure may include at least one of steps S2301 to S2305. For example, step S2303 may be implemented as an independent embodiment, step S2304 may be implemented as an independent embodiment, and step S2305 may be implemented as an independent embodiment, but is not limited thereto.

[0240] In some embodiments, steps S2303 and S2304 may be performed in an alternate order or simultaneously.

[0241] In some embodiments, steps S2301 and S2302 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0242] In some embodiments, steps S2304 and S2305 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0243] In some embodiments, other alternative implementations may be described before or after the specification corresponding to FIG2C.

[0244] In some embodiments, the names of information, etc., are not limited to the names described in the embodiments. Terms such as "information", "message", "signal", "signaling", "report", "configuration", "indication", "instruction", "command", "channel", "parameter", "domain", "field", "symbol", "symbol", "codebook", "codeword", "codepoint", "bit", "data", "program", and "chip" can be used interchangeably.

[0245] In some embodiments, the terms “downlink control information (DCI),” “downlink (DL) assignment,” “DL DCI,” “uplink (UL) grant,” and “UL DCI” can be used interchangeably.

[0246] In some embodiments, terms such as "physical downlink shared channel (PDSCH)" and "DL data" can be used interchangeably, as can terms such as "physical uplink shared channel (PUSCH)" and "UL data".

[0247] In some embodiments, terms such as “moment,” “point in time,” “time,” and “time location” can be used interchangeably, as can terms such as “duration,” “segment,” “time window,” “window,” and “time.”

[0248] In some embodiments, the terms "precoding", "precoder", "weight", "precoding weight", "quasi-co-location (QCL)", "transmission configuration indication (TCI) status", "spatial relation", "spatial domain filter", "transmission power", "phase rotation", "antenna port", "antenna port group", "layer", "the number of layers", "rank", "resource", "resource set", "resource group", "beam", "beam width", "beam angular degree", "antenna", "antenna element", and "panel" can be used interchangeably.

[0249] In some embodiments, “get,” “obtain,” “receive,” “transmit,” “bidirectional transmission,” and “send and / or receive” can be used interchangeably and can be interpreted as receiving from other entities, obtaining from protocols, obtaining from higher layers, obtaining through self-processing, or autonomous implementation, among other meanings.

[0250] In some embodiments, terms such as “send,” “transmit,” “report,” “distribute,” “transfer,” “bidirectional transmission,” “send and / or receive” can be used interchangeably.

[0251] In some embodiments, terms such as "certain," "preset," "default," "set," "indicated," "a certain," "any," and "first" can be used interchangeably. "Certain A," "preset A," "default A," "set A," "indicated A," "a certain A," "any A," and "first A" can be interpreted as A pre-defined in a protocol or the like, or as A obtained through setting, configuration, or instruction, or as specific A, a certain A, any A, or first A, but are not limited thereto.

[0252] In some embodiments, the determination or judgment can be made by a value represented by 1 bit (0 or 1), or by a true or false value (Boolean value (bool)) represented by true or false, or by a numerical comparison (e.g., a comparison with a predetermined value), but is not limited thereto.

[0253] In some embodiments, "not expecting to receive" can be interpreted as not receiving on time domain resources and / or frequency domain resources, or as not performing subsequent processing on the data after receiving it; "not expecting to send" can be interpreted as not sending, or as sending but not expecting the receiver to respond to the sent content.

[0254] Figure 3 is a flowchart illustrating a model transmission method according to an embodiment of the present disclosure. As shown in Figure 3, this embodiment relates to a model transmission method executed by a first device, the method comprising:

[0255] Step S3101: Send the first information to the second device.

[0256] The optional implementation of step S3101 can be found in the optional implementation of step S2101 in Figure 2A and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0257] Step S3102: Send the third information to the second device.

[0258] The optional implementation of step S3102 can be found in the optional implementation of step S2102 in Figure 2A and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0259] Step S3103: Receive the first message sent by the second device.

[0260] The optional implementation of step S3103 can be found in the optional implementation of step S2103 in Figure 2A and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0261] The model transfer method involved in the embodiments of this disclosure may include at least one of steps S3101 to S3103. For example, step S3101 may be implemented as an independent embodiment, step S3102 may be implemented as an independent embodiment, and step S3103 may be implemented as an independent embodiment, but is not limited thereto.

[0262] In some embodiments, step S3102 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0263] In some embodiments, step S3103 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0264] Figure 4 is a flowchart illustrating a model transmission method according to an embodiment of the present disclosure. As shown in Figure 4, this embodiment relates to a model transmission method executed by a second device, the method comprising:

[0265] Step S4101: Receive the first information sent by the first device.

[0266] The optional implementation of step S4101 can be found in the optional implementation of step S2101 in Figure 2A and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0267] Step S4102: Receive the third information sent by the first device.

[0268] The optional implementation of step S4102 can be found in the optional implementation of step S2102 in Figure 2A and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0269] Step S4103: Send the first message to the first device.

[0270] The optional implementation of step S4103 can be found in the optional implementation of step S2103 in Figure 2A and other related parts in the embodiments involved in Figure 2A, which will not be repeated here.

[0271] The model transfer method involved in the embodiments of this disclosure may include at least one of steps S4101 to S4103. For example, step S4101 may be implemented as an independent embodiment, step S4102 may be implemented as an independent embodiment, and step S4103 may be implemented as an independent embodiment, but is not limited thereto.

[0272] In some embodiments, step S4102 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0273] In some embodiments, step S4103 is optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0274] In some embodiments, the above methods may include the methods of the embodiments described above on the communication system side, the first device side, the network device side, etc., which will not be repeated here.

[0275] This disclosure proposes a method for transmitting AI models, which utilizes the correlation between the content of the models to reduce the amount of data transmitted by the models, thereby effectively improving the utilization of wireless resources and reducing the latency of AI model transmission.

[0276] In this embodiment of the disclosure, when updating an AI model, if the model structure remains unchanged, an incremental transmission method is adopted, that is, only the parts that are different from the AI ​​model on which it is based (i.e., the baseline model) are transmitted.

[0277] Example 1: The access network device sends the AI ​​model, and the terminal receives the AI ​​model. Alternatively, the terminal sends the AI ​​model, and the access network device receives the AI ​​model.

[0278] Option 1: When updating an AI model, a list is provided, where each element contains the following information: the index / position information in the AI ​​model data, the number of bits changed, and the updated value. In this option, the AI ​​model is treated as a byte stream, and updating the AI ​​model indicates the changed position in this byte stream and the changed value.

[0279] Example 2: Taking the model structure shown in Figure 2D as an example. The input and output of this model are X and Y, respectively, both of which are vectors (one-dimensional tensors) with two elements. The operation performed is Y = Add(MatMul(A,X),C), that is, Y = AX + C, where MatMul(A,X) is the matrix product of A and X, and Add(AX,C) is the sum of AX and C. The parameters of this model include matrix A and vector C.

[0280] Assuming all data types in the model are 32-bit floating-point numbers, If the model is stored in ONNX format, the model size is 140 bytes, and the model content is as follows:

[0281] The above is displayed in hexadecimal. Hexadecimal 0x08 is the first byte of the model content, 0x09 is the second byte, 0x3a is the third byte, and so on.

[0282] If the AI ​​model is updated, but the model structure remains unchanged, the model parameters (Bold numbers indicate changed parameters). The updated AI model content is shown below, still 140 bytes, with changes relative to the baseline model indicated by underlines.

[0283] Using Scheme 1, when updating the AI ​​model, a list containing two elements is used. The first element indicates (0x43, 0x04, [0x9a, 0x99, 0x19, 0x3f]), meaning that the four bytes of data starting from position 0x43 in the AI ​​model data (where the first byte of the AI ​​model is at position 0) are updated to [0x9a, 0x99, 0x19, 0x3f]. In this scheme, hexadecimal is used to ensure the AI ​​model content corresponds correctly. Those skilled in the art can also use decimal or other bases. The second element indicates (0x5c, 0x02, [0x4c, 0x3e]), meaning that the two bytes of data starting from position 0x5c in the AI ​​model data are updated to [0x4c, 0x3e]. Therefore, a total of 10 bytes are required (assuming that both the position information and the number of changed bits are indicated by one byte). Compared to using 140 bytes, this significantly reduces the amount of data transmitted.

[0284] Option 2: When updating an AI model, perform an exclusive OR operation between the content of the baseline model and the content of the updated model, and then compress the XOR result before transmission. The receiver decompresses the received content and then XORs it with the baseline model content to obtain the updated model content. Similar to Option 1, this option treats the AI ​​model as a bit stream. For unchanged parts, the XOR operation results in 0. Numerous repeated 0s are compressed, resulting in a significant reduction in data volume.

[0285] Option 2.1: When the model structure and model parameters of the AI ​​model can be distinguished at the content level of transmission (for example, in the model file, the first part is the model structure and the second part is the model parameters), then only the content of the model parameters of the baseline model and the content of the model parameters of the updated model are XORed and compressed before transmission.

[0286] Example 3: Perform an XOR operation between the content of the baseline model in Example 2 and the content of the updated model. The result is as follows:

[0287] As shown, after compression using gzip (gzip is a data compression format based on the DEFLATE algorithm, and DEFLATE is a lossless data compression algorithm), the data size is 43 bytes (including some header information, such as a 2-byte magic number: 0x1f, 0x8b, and the compression algorithm: 0x08 indicating DEFLATE). Compared to using 140 bytes, this significantly reduces the amount of data transmitted.

[0288] The result of XORing the content of the baseline model with the content of the updated model is as follows:

[0289] Option 3: When updating an AI model, a list is provided, where each element corresponds to a changed parameter in the model, containing at least the following information: parameter index and parameter value, where the parameter index indicates the position of the changed parameter in the model data. Unlike Options 1 and 2, this option indicates the changed parameters at the level of the AI ​​model's content. In this option, a parameter can refer to the smallest data unit in the AI ​​model, such as a 2×2 matrix containing 4 elements, each of which can be considered a parameter.

[0290] Option 3.1: The parameter value is the parameter value used by the updated AI model, which is an "absolute" value.

[0291] Option 3.2: The parameter values ​​are the changes in the updated AI model relative to the baseline version of the AI ​​model. When using changes, fewer bits can be used to indicate the parameter values ​​(compared to Option 3.1).

[0292] Scheme 3.3: The parameter index (the position of the parameter in the model data) refers to the index / position of the parameter among all the parameters transmitted in the model data.

[0293] Option 3.4: When the absolute value (mathematical absolute value) of the change in a parameter value is less than or equal to a threshold, the list does not include information about the parameter (parameter index and parameter value) corresponding to that parameter value. This further reduces the amount of data transmitted for the AI ​​model.

[0294] Solution 3.4.1: When the terminal sends the AI ​​model, the threshold corresponding to the change in the network device configuration parameter value.

[0295] Example 4: Taking the baseline model from Example 2 and the updated model as examples. Updated model parameters. (Bold numbers indicate changed parameters). Taking the smallest data unit in the AI ​​model as the parameter, the model contains 6 parameters. The parameter indices (according to scheme 3.3) are 0, 1, 2, 3, 4, 5, and the corresponding parameter values ​​are 0.7, -0.9, 0.6, 1.2, 0.8, and 0.2, respectively.

[0296] As per scheme 3.1, a list is specified, including the following elements: (2, 0.6), (5, 0.2), that is, the parameter at index 2 is updated to 0.6, and the parameter at index 5 is updated to 0.2.

[0297] As per scheme 3.2, a list is provided containing the following elements: (2, 0.2), (5, 0.3), where the parameter at index 2 changes to 0.6 - 0.4 = 0.2, and the parameter at index 5 changes to 0.2 - (-0.1) = 0.3.

[0298] According to scheme 3.4, the threshold is set to 0.2. Since the parameter change value of 0.2 at index 2 is within the threshold, the list does not include the relevant information of the parameter corresponding to that parameter value. According to scheme 3.1, the list only contains the element (5, 0.2); according to scheme 3.2, the list only contains the element (5, 0.3).

[0299] Option 4: When transmitting the AI ​​model, indicate whether to use the incremental transmission method described in the main invention point. If the incremental transmission method is not used, then transmit all parameter information of the AI ​​model.

[0300] Option 4.1: When transmitting the AI ​​model, indicate the version number (ID) of the current model, and optionally indicate the version number of the model on which the current model is based. If the version number of the model on which the current model is based is indicated, it is considered as indicating that an incremental transmission method is used, that is, the AI ​​model corresponding to the version number of the model on which the current model is based is the aforementioned base model. If the version number of the model on which the current model is based is not indicated, it is considered as indicating that an incremental transmission method is not used. The base model can be a model that has been communicated offline, a related AI model from the most recent transmission, or a related model from the most recent transmission whose transmission has been confirmed by the recipient.

[0301] Option 5: When the terminal receives an AI model using incremental transmission, the terminal directly updates the basic model based on the content of the incremental transmission.

[0302] Option 6: When the terminal receives an AI model using incremental transmission, it updates a copy of the base model based on the incremental transmission content. If the terminal fails to update the AI ​​model, it falls back to the base model or the most recently used AI model and reports the update failure to the network device. The reported information may include the AI ​​model's identifier, the base model's version number, and the updated model's version number.

[0303] Option 7: The network device sends information to the terminal to configure whether the terminal uses incremental transmission when transmitting AI models uplink.

[0304] Option 8: The terminal notifies the network via UE capability signaling whether it supports incremental transmission when transmitting AI models. The UE capability can separately indicate whether it supports receiving AI model transmissions using incremental transmission in the downlink and whether it supports transmitting AI models using incremental transmission in the uplink. The UE capability can also provide joint indications for downlink and uplink.

[0305] In the embodiments disclosed herein, some or all of the steps and their optional implementations may be arbitrarily combined with some or all of the steps in other embodiments, or may be arbitrarily combined with the optional implementations in other embodiments.

[0306] This disclosure also provides an apparatus for implementing any of the above methods. For example, an apparatus is provided that includes units or modules for implementing the steps performed by the first device in any of the above methods. Alternatively, another apparatus is provided that includes units or modules for implementing the steps performed by a network device (e.g., an access network device, a core network functional node, a core network device, etc.) in any of the above methods.

[0307] It should be understood that the division of units or modules in the above device is only a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, the units or modules in the device can be implemented by a processor calling software: for example, the device includes a processor connected to a memory containing instructions. The processor calls the instructions stored in the memory to implement any of the above methods or to implement the functions of the units or modules in the above device. The processor can be, for example, a general-purpose processor, such as a Central Processing Unit (CPU) or a microprocessor, and the memory can be internal or external to the device. Alternatively, the units or modules in the device can be implemented in the form of hardware circuits. The functionality of some or all of the units or modules can be achieved through the design of these hardware circuits, which can be understood as one or more processors. For example, in one implementation, the hardware circuit is an application-specific integrated circuit (ASIC). The functionality of some or all of the units or modules is achieved through the design of the logical relationships between the components within the circuit. In another implementation, the hardware circuit can be implemented using a programmable logic device (PLD). Taking a field-programmable gate array (FPGA) as an example, it can include a large number of logic gates. The connection relationships between the logic gates are configured through configuration files, thereby achieving the functionality of some or all of the units or modules. All units or modules of the above device can be implemented entirely through processor-called software, entirely through hardware circuits, or partially through processor-called software with the remaining parts implemented through hardware circuits.

[0308] In this embodiment, the processor is a circuit with signal processing capabilities. In one implementation, the processor can be a circuit with instruction read and execute capabilities, such as a Central Processing Unit (CPU), a microprocessor, a graphics processing unit (GPU) (which can be understood as a microprocessor), or a digital signal processor (DSP). In another implementation, the processor can implement certain functions through the logical relationships of hardware circuits. The logical relationships of the aforementioned hardware circuits are fixed or reconfigurable. For example, the processor is a hardware circuit implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD), such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document and configuring the hardware circuit can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units or modules. Furthermore, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as a Neural Network Processing Unit (NPU), a Tensor Processing Unit (TPU), or a Deep Learning Processing Unit (DPU).

[0309] Figure 5A is a schematic diagram of the structure of the first device proposed in an embodiment of this disclosure. As shown in Figure 5A, the first device 5100 may include a transceiver module 5101. In some embodiments, the transceiver module 5101 sends first information to a second device. Optionally, the transceiver module is used to perform at least one of the transceiver steps (such as step S2101, but not limited thereto) performed by the first device in any of the above methods, which will not be described in detail here.

[0310] In some embodiments, the first device may further include a processing module.

[0311] In some embodiments, the AI ​​model is based on a bitstream or bytestream representation; the first information includes at least one of the following: the position information of the change in the bitstream or the bytestream; the number of bits that have changed in the bitstream; the number of bytes that have changed in the bytestream; and the changed value.

[0312] In some embodiments, the processing module is used to compress the second information to obtain the first information, wherein the second information is obtained by performing an XOR operation on the model parameters before the AI ​​model change and the model parameters after the change.

[0313] In some embodiments, the content of the AI ​​model includes model structure data and model parameters. Without changing the model structure data, the second information is compressed to obtain the first information.

[0314] In some embodiments, the first information includes at least one of the following: the parameter index of the first parameter; the parameter value of the first parameter.

[0315] In some embodiments, the parameter value is the value after the first parameter has been changed, or the parameter value is the change value of the first parameter.

[0316] In some embodiments, the parameter values ​​included in the first information are greater than a threshold value.

[0317] In some embodiments, the transceiver module is further configured to send third information to the second device, the third information being used to indicate the use of incremental transmission mode.

[0318] In some embodiments, the third information includes the current version number and the baseline version number corresponding to the AI ​​model.

[0319] In some embodiments, the first information is used by the second device to update the AI ​​model.

[0320] In some embodiments, the transceiver module is further configured to receive a first message sent by the second device, the first message being used to indicate that the model update has failed, the first message including at least one of the following: the identifier of the AI ​​model, the current version number, and the baseline version number.

[0321] In some embodiments, where the first device is a terminal device and the second device is a network device, the transceiver module is further configured to: send fourth information to the network device, the fourth information being used to indicate that the terminal device supports uplink transmission using an incremental transmission method; and receive fifth information sent by the network device, the fifth information being used to configure the terminal device to perform uplink transmission using an incremental transmission method.

[0322] In some embodiments, where the first device is a network device and the second device is a terminal device, the transceiver module is further configured to: receive sixth information sent by the terminal device, the sixth information being used to indicate that the terminal device supports downlink reception using incremental transmission mode; and send seventh information to the terminal device, the seventh information being used to configure the terminal to perform downlink reception using incremental transmission mode.

[0323] Figure 5B is a schematic diagram of the structure of the second device proposed in an embodiment of this disclosure. As shown in Figure 5B, the second device 5200 may include a transceiver module 5201. In some embodiments, the transceiver module 5201 is used to receive first information sent by the first device. Optionally, the transceiver module is used to perform at least one of the sending and receiving steps performed by the second device in any of the above methods, which will not be described in detail here.

[0324] In some embodiments, the second device may further include a processing module.

[0325] In some embodiments, the AI ​​model is based on a bitstream or bytestream representation; the first information includes at least one of the following: the position information of the change in the bitstream or the bytestream; the number of bits that have changed in the bitstream; the number of bytes that have changed in the bytestream; and the changed value.

[0326] In some embodiments, the first information is obtained by compressing the second information, and the second information is obtained by performing an XOR operation on the model parameters before and after the change.

[0327] In some embodiments, the content of the AI ​​model includes model structure data and model parameters. Without changing the model structure data, the second information is compressed to obtain the first information.

[0328] In some embodiments, the first information includes at least one of the following: the parameter index of the first parameter; the parameter value of the first parameter.

[0329] In some embodiments, the parameter value is the value after the first parameter has been changed, or the parameter value is the change value of the first parameter.

[0330] In some embodiments, the parameter values ​​included in the first information are greater than a threshold value.

[0331] In some embodiments, the transceiver module is further configured to receive third information sent by the first device, the third information being used to indicate the use of incremental transmission mode.

[0332] In some embodiments, the third information includes the current version number and the baseline version number corresponding to the AI ​​model.

[0333] In some embodiments, the first information is used by the second device to update the AI ​​model.

[0334] In some embodiments, the transceiver module is further configured to send a first message to the first device, the first message indicating that the model update has failed, the first message including at least one of the following: the identifier of the AI ​​model, the current version number, and the baseline version number.

[0335] In some embodiments, where the first device is a terminal device and the second device is a network device, the transceiver module is further configured to: receive fourth information sent by the terminal device, the fourth information being used to indicate that the terminal device supports uplink transmission using an incremental transmission method; and send fifth information to the terminal device, the fifth information being used to configure the terminal to use an incremental transmission method for uplink transmission.

[0336] In some embodiments, where the first device is a network device and the second device is a terminal device, the transceiver module is further configured to: send a sixth message to the network device, the sixth message being used to instruct the terminal device to support downlink reception using an incremental transmission method; and receive a seventh message sent by the network device, the seventh message being used to configure the terminal to perform downlink reception using an incremental transmission method.

[0337] Figure 6A is a schematic diagram of the structure of the communication device 6100 proposed in an embodiment of this disclosure. The communication device 6100 can be a network device (e.g., access network device, core network device, etc.), a terminal (e.g., user equipment, etc.), a chip, chip system, or processor that supports the network device in implementing any of the above methods, or a chip, chip system, or processor that supports the terminal in implementing any of the above methods. The communication device 6100 can be used to implement the methods described in the above method embodiments; for details, please refer to the descriptions in the above method embodiments.

[0338] As shown in Figure 6A, the communication device 6100 includes one or more processors 6101. The processor 6101 can be a general-purpose processor or a dedicated processor, such as a baseband processor or a central processing unit (CPU). The baseband processor can be used to process communication protocols and communication data, while the CPU can be used to control communication devices (e.g., base stations, baseband chips, terminal devices, terminal device chips, DUs or CUs, etc.), execute programs, and process program data. Optionally, the communication device 6100 can be used to execute any of the above methods. Optionally, one or more processors 6101 can be used to invoke instructions to cause the communication device 6100 to execute any of the above methods.

[0339] In some embodiments, the communication device 6100 further includes one or more transceivers 6102. When the communication device 6100 includes one or more transceivers 6102, the transceiver 6102 performs at least one of the communication steps such as sending and / or receiving in the above method (e.g., step S2101, but not limited thereto), and the processor 6101 performs at least one of the other steps. In optional embodiments, the transceiver may include a receiver and / or a transmitter, which may be separate or integrated together. Optionally, the terms transceiver, transceiver unit, transceiver, transceiver circuit, interface circuit, interface, etc., can be used interchangeably; the terms transmitter, sending unit, transmitter, sending circuit, etc., can be used interchangeably; and the terms receiver, receiving unit, receiver, receiving circuit, etc., can be used interchangeably.

[0340] In some embodiments, the communication device 6100 further includes one or more memories 6103 for storing data. Optionally, all or part of the memories 6103 may be located outside the communication device 6100. In optional embodiments, the communication device 6100 may include one or more interface circuits 6104. Optionally, the interface circuits 6104 are connected to the memories 6103 and can be used to receive data from the memories 6103 or other devices, and to send data to the memories 6103 or other devices. For example, the interface circuits 6104 can read data stored in the memories 6103 and send that data to the processor 6101.

[0341] The communication device 6100 described in the above embodiments may be a network device or a terminal, but the scope of the communication device 6100 described in this disclosure is not limited thereto, and the structure of the communication device 6100 may not be limited by FIG. 6A. The communication device may be a standalone device or a part of a larger device. For example, the communication device may be: (1) a standalone integrated circuit IC, or chip, or chip system or subsystem; (2) a collection of one or more ICs, optionally, the IC collection may also include storage components for storing data and programs; (3) an ASIC, such as a modem; (4) a module that can be embedded in other devices; (5) a receiver, terminal device, smart terminal device, cellular phone, wireless device, handheld device, mobile unit, vehicle device, network device, cloud device, artificial intelligence device, etc.; (6) others, etc.

[0342] Figure 6B is a schematic diagram of the structure of chip 6200 according to an embodiment of this disclosure. For cases where the communication device 6100 can be a chip or a chip system, please refer to the schematic diagram of chip 6200 shown in Figure 6B, but it is not limited thereto.

[0343] Chip 6200 includes one or more processors 6201. Chip 6200 is used to perform any of the methods described above.

[0344] In some embodiments, chip 6200 further includes one or more interface circuits 6202. Optionally, terms such as interface circuit, interface, and transceiver pin can be used interchangeably. In some embodiments, chip 6200 further includes one or more memories 6203 for storing data. Optionally, all or part of the memories 6203 may be located outside chip 6200. Optionally, interface circuit 6202 is connected to memory 6203, and interface circuit 6202 can be used to receive data from memory 6203 or other devices, and interface circuit 6202 can be used to send data to memory 6203 or other devices. For example, interface circuit 6202 can read data stored in memory 6203 and send the data to processor 6201.

[0345] In some embodiments, the interface circuit 6202 performs at least one of the communication steps such as sending and / or receiving in the above-described method (e.g., step S2101, but not limited thereto). For example, the interface circuit 6202 performing the communication steps such as sending and / or receiving in the above-described method means that the interface circuit 6202 performs data interaction between the processor 6201, the chip 6200, the memory 6203, or the transceiver device. In some embodiments, the processor 6201 performs at least one of the other steps.

[0346] The modules and / or devices described in the various embodiments, such as virtual devices, physical devices, and chips, can be combined or separated arbitrarily as needed. Optionally, some or all steps can also be performed collaboratively by multiple modules and / or devices, which is not limited here.

[0347] This disclosure also proposes a storage medium storing instructions that, when executed on the communication device 6100, cause the communication device 6100 to perform any of the above methods. Optionally, the storage medium is an electronic storage medium. Optionally, the storage medium is a computer-readable storage medium, but not limited thereto; it may also be a storage medium readable by other devices. Optionally, the storage medium may be a non-transitory storage medium, but not limited thereto; it may also be a temporary storage medium.

[0348] This disclosure also provides a program product that, when executed by the communication device 6100, causes the communication device 6100 to perform any of the above methods. Optionally, the program product is a computer program product.

[0349] This disclosure also proposes a computer program that, when run on a computer, causes the computer to perform any of the above methods.

Claims

1. A model transfer method, characterized in that, Performed by a first device, the method includes: Send a first message to the second device, the first message being used to indicate a first parameter of the model parameters of the artificial intelligence (AI) model that has changed.

2. The method according to claim 1, characterized in that, The AI ​​model is based on bitstream or bytestream representation; The first information includes at least one of the following: The location information of the change in the bit stream or the byte stream; The number of bits that changed in the bit stream; The number of bytes that changed in the byte stream; The changed values.

3. The method according to claim 1, characterized in that, The method further includes: The first information is obtained by compressing the second information. The second information is obtained by performing an XOR operation on the model parameters before the AI ​​model was changed and the model parameters after the change.

4. The method according to claim 3, characterized in that, The AI ​​model includes model structure data and model parameters. Without changing the model structure data, the second information is compressed to obtain the first information.

5. The method according to claim 1, characterized in that, The first information includes at least one of the following: The parameter index of the first parameter; The parameter value of the first parameter.

6. The method according to claim 5, characterized in that, The parameter value is either the value after the first parameter has been changed, or the parameter value is the change in the first parameter.

7. The method according to claim 5 or 6, characterized in that, The parameter values ​​included in the first information are greater than the threshold value.

8. The method according to claim 1, characterized in that, The method further includes: Send a third message to the second device, the third message being used to indicate the use of incremental transmission mode.

9. The method according to claim 8, characterized in that, The third piece of information includes the current version number and the baseline version number corresponding to the AI ​​model.

10. The method according to any one of claims 1 to 9, characterized in that, The first information is used by the second device to update the AI ​​model.

11. The method according to claim 10, characterized in that, The method further includes: The system receives a first message sent by the second device, the first message indicating that the model update has failed, and the first message includes at least one of the following: the identifier of the AI ​​model, the current version number, and the base version number.

12. The method according to any one of claims 1 to 11, characterized in that, When the first device is a terminal device and the second device is a network device, the method further includes at least one of the following: Send a fourth message to the network device, the fourth message being used to instruct the terminal device to support uplink transmission using incremental transmission mode; The terminal device receives a fifth message sent by the network device, the fifth message being used to configure the terminal device to use an incremental transmission method for uplink transmission.

13. The method according to any one of claims 1 to 11, characterized in that, When the first device is a network device and the second device is a terminal device, the method further includes at least one of the following: The terminal device receives a sixth message, which indicates that the terminal device supports downlink reception using incremental transmission. A seventh message is sent to the terminal device, the seventh message being used to configure the terminal device to use incremental transmission mode for downlink reception.

14. A model transfer method, characterized in that, Performed by a second device, the method includes: Receive first information sent by a first device, the first information being used to determine a first parameter indicating that the model parameters of the artificial intelligence (AI) model have changed.

15. The method according to claim 14, characterized in that, The AI ​​model is based on bitstream or bytestream representation; The first information includes at least one of the following: The location information of the change in the bit stream or the byte stream; The number of bits that changed in the bit stream; The number of bytes that changed in the byte stream; The changed values.

16. The method according to claim 14, characterized in that, The first information is obtained by compressing the second information, which is obtained by performing an XOR operation on the model parameters before and after the change.

17. The method according to claim 16, characterized in that, The AI ​​model includes model structure data and model parameters. Without changing the model structure data, the second information is compressed to obtain the first information.

18. The method according to claim 14, characterized in that, The first information includes at least one of the following: The parameter index of the first parameter; The parameter value of the first parameter.

19. The method according to claim 18, characterized in that, The parameter value is either the value after the first parameter has been changed, or the parameter value is the change in the first parameter.

20. The method according to claim 18 or 19, characterized in that, The parameter values ​​included in the first information are greater than the threshold value.

21. The method according to claim 14, characterized in that, The method further includes: The system receives third information sent by the first device, the third information being used to indicate the use of incremental transmission mode.

22. The method according to claim 21, characterized in that, The third piece of information includes the current version number and the baseline version number corresponding to the AI ​​model.

23. The method according to any one of claims 14 to 22, characterized in that, The first information is used by the second device to update the AI ​​model.

24. The method according to claim 23, characterized in that, The method further includes: A first message is sent to the first device, the first message being used to indicate that the model update has failed, the first message including at least one of the following: the identifier of the AI ​​model, the current version number, and the baseline version number.

25. The method according to any one of claims 14 to 24, characterized in that, When the first device is a terminal device and the second device is a network device, the method further includes at least one of the following: The terminal device receives a fourth message, which indicates that the terminal device supports uplink transmission using an incremental transmission method. A fifth message is sent to the terminal device, the fifth message being used to configure the terminal to use an incremental transmission method for uplink transmission.

26. The method according to any one of claims 14 to 24, characterized in that, When the first device is a network device and the second device is a terminal device, the method further includes at least one of the following: Send a sixth message to the network device, the sixth message being used to instruct the terminal device to support downlink reception using incremental transmission mode; The terminal receives a seventh message sent by the network device, the seventh message being used to configure the terminal to use incremental transmission mode for downlink reception.

27. A communication device, characterized in that, The communication device is used to perform the model transmission method according to any one of claims 1 to 13 or the model transmission method according to any one of claims 14 to 26.

28. A communication system, characterized in that, The device includes a first device and a second device, wherein the first device is configured to implement the model transmission method of any one of claims 1 to 13, and the second device is configured to implement the model transmission method of any one of claims 14 to 26.

29. A storage medium, characterized in that, The storage medium stores instructions that, when executed on a communication device, cause the communication device to perform the model transmission method as described in any one of claims 1 to 13 or the model transmission method as described in any one of claims 14 to 26.

30. A program product, characterized in that, It includes at least one of a program and instructions, wherein when the program or instructions are executed by a communication device, they implement the model transmission method of any one of claims 1 to 13 or perform the model transmission method of any one of claims 14 to 26.

Images