CN117764146A

CN117764146A - Information transmission method, device and system

Info

Publication number: CN117764146A
Application number: CN202211176487.XA
Authority: CN
Inventors: 郭子阳; 刘鹏; 颜敏; 董明杰; 杨讯; 罗嘉俊
Original assignee: Huawei Technologies Co Ltd
Current assignee: Huawei Technologies Co Ltd
Priority date: 2022-09-26
Filing date: 2022-09-26
Publication date: 2024-03-26
Also published as: WO2024067143A1

Abstract

The embodiment of the application provides an information transmission method, an information transmission device and an information transmission system, which are used for improving the flexibility of a training method of a model and relate to the technical field of wireless communication. In the method, the first device may indicate to the second device a dimension of the status information to be transmitted, and the second device transmits to the first device the status information conforming to the dimension. Based on the scheme, different second devices can be allowed to use different state information, and meanwhile, the flexibility of the model training method can be improved because the dimension of the state information indicated by the first device is smaller than that of the original state information. In the method, the first device may determine a first model for performing a task based on the state information transmitted by the second device, and transmit the first model to the second device. Since it is determined that the dimension of the state information of the first model is smaller than that of the original state information, the parameter amount of the first model is also smaller than that of the model for performing the task in the related art, and overhead can be effectively transmitted.

Description

Information transmission method, device and system

Technical Field

The present disclosure relates to the field of wireless communications technologies, and in particular, to an information transmission method, apparatus, and system.

Background

New wireless technologies, new terminals and new applications will make wireless networks more and more complex. To combat the trend of high complexity of wireless networks, artificial intelligence (artificial intelligence, AI) has become a common practice in the industry as an effective tool for wireless network design and management.

Existing AI algorithms can be divided into two categories, distributed training distribution execution and center training distribution execution. The central training service distributed execution means that the central node trains the neural network, and the parameters of the neural network are issued to the distributed nodes, and the distributed nodes conduct prediction or decision through the neural network. Distributed training distribution execution refers to training a neural network by distributed nodes and making predictions or decisions by the distributed nodes through the neural network. The central training profile execution may utilize global information, and thus may result in greater performance gains than the distributed training profile execution.

However, in the scheme of central training distribution execution at present, the status information reported by each distributed node, namely, the format of the input information of the neural network, which information the status information specifically contains, the dimension and the arrangement sequence of each status information, and the like, are required to be unified, so that personalized algorithm design of each manufacturer is inconvenient, and the training method of the model lacks flexibility.

Disclosure of Invention

The embodiment of the application provides an information transmission method, device and system, which are used for improving the flexibility of a training method of a model.

In a first aspect, an information transmission method is provided. The method may be performed by the first device or by a chip/chip system. In the method, a first device sends first information to a second device. Wherein the first information indicates a dimension of first state information, the first state information being input information of the first model. In the above, the first model comprises a neural network for assisting the second device in making predictions or decisions. The first device receives second information from the second device, the second information including first state information, a dimension of the first state information conforming to a dimension indicated by the first information. The first device determines a first model based on the first state information. The first device sends third information to the second device, the third information including parameters of the first model, the parameters of the first model including at least one of bias and weight.

Based on the scheme, different distributed nodes can be allowed to use different state information, and meanwhile, the dimension of the state information indicated by the central node is smaller than that of original state information sent by the distributed node to the central node in the related technology, so that the overhead of reporting data is effectively reduced. In the method, the central node may determine a first model for performing a task based on the state information transmitted by the distributed node, and transmit the first model to the distributed node. Since the dimension of the state information of the first model is determined to be smaller than the dimension of the original state information used for training the model in the related art, the parameter quantity of the first model is also smaller than the parameter quantity of the model used for executing the task in the related art, and the cost for sending the model can be effectively reduced. In addition, the method does not need the distributed nodes to use the same state information, and can have a good privacy protection effect on the structure and parameters of the model.

In the above, the first model is related to the task performed by the second device. In one possible implementation, a first device may receive a first frame from a second device, which may be used to instruct the second device to perform a task, or the first frame may be used to instruct the second device to request an AI-assisted function/task to be turned on.

In one possible implementation, the third information further includes information indicating a structure of the first model. In a possible case, the information indicating the structure of the first model includes the number of layers of the first model, the type of neural network of each layer of the first model, the number of neurons, and the activation function. In another possible case, the information indicating the structure of the first model includes a model index.

Based on the scheme, the first device can indicate the first model to the second device through the third information, so that the second device can adopt the first model to assist decision making or prediction.

In one possible implementation, the information for indicating the structure of the first model further includes jump connection indicating information, where the jump connection indicating information includes a start layer of the first model and an end layer of the first model, and the jump connection indicating information is used to indicate jump connection of the start layer and the end layer. Optionally, the hop connection indication information may further include a hop connection number for indicating that several hop connections exist in the first model.

Based on the above scheme, the first device may indicate to the second device whether the neural network included in the first model contains a hopping connection. The jump connection structure is favorable for the fusion of the bottom layer characteristics and the high layer characteristics, and can improve the learning ability of the neural network.

In one possible implementation, the third information may further include fourth information. The fourth information indicates whether the first model contains a jump connection or whether the information indicating the structure of the first model contains the jump connection indicating information above. If the fourth information indicates that the first model does not contain the jump connection, jump connection indication information does not exist in the information for indicating the structure of the first model, so that overhead can be saved, and processing resources of the second device can be saved.

In one possible implementation, the second information further includes an action including an operation performed by the first device and a reward for describing an evaluation of the operation performed by the first device. Wherein the actions and rewards may be used for reinforcement learning of the first model. Based on the scheme, through actions and rewards, the accuracy of prediction or decision of the first model can be improved.

In one possible implementation, the second information further includes a tag for indicating the desired output information of the first model. Based on the scheme, the accuracy of the prediction or decision of the first model can be improved through the label.

In a possible implementation manner, the second information further includes task information, where the task information is used to indicate a task corresponding to the first state information. Based on the scheme, the second device can indicate the task corresponding to the first state information to the first device, so that the first device can train the first model of the corresponding task according to the first state information.

In a second aspect, an information transmission method is provided. The method may be performed by the second device or by a chip/chip system. In the method, a second device receives first information from a first device, the first information indicating a dimension of first state information, the first state information being input information of a first model. The first model includes a neural network for assisting the second device in making predictions or decisions. The second device sends second information to the first device, the second information comprising first state information, the dimensions of the first state information conforming to the dimensions indicated by the first information. The second device receives parameters of the first model from the first device. The parameters of the first model include at least one of bias and weight. The second device inputs the first state information into the first model to obtain output information of the first model. The output information may be used for a second device decision or prediction.

In the above, the first model is related to the task performed by the second device. In one possible implementation, the second device may send a first frame to the first device, which may be used to instruct the second device to perform a task, or the first frame may be used to instruct the second device to request an AI-assisted function/task to be turned on.

In one possible implementation, the second information further includes an action including an operation performed by the first device and a reward for describing an evaluation of the operation performed by the first device. Wherein the actions and rewards may be used for reinforcement learning of the first model.

In one possible implementation, the second information further includes a tag for indicating the desired output information of the first model.

In a possible implementation manner, the second information further includes task information, where the task information is used to indicate a task corresponding to the first state information.

In one possible implementation, the second information is derived from a second model, which is trained by: the second state information is used as the input of the second model, and the output of the second model is the first state information. The dimension of the second state information is greater than the dimension of the first state information. And taking the first state information as the input of a third model, wherein the output of the third model is third state information, the dimension of the third state information is larger than that of the first state information, and the dimension of the third state information is equal to that of the second state information. The second model is trained by minimizing errors in the third state information and the second state information. Based on the scheme, the second device can determine the first state information sent to the first device through the second model, and the efficiency and accuracy of acquiring the first state information can be improved.

In a third aspect, a communication device is provided, comprising a processing unit and a transceiver unit.

And the receiving and transmitting unit is used for transmitting first information to the second equipment, wherein the first information indicates the dimension of the first state information, and the first state information is input information of the first model. The first model includes a neural network for assisting the second device in making predictions or decisions. The receiving and transmitting unit is further used for receiving second information from the second device, the second information comprises first state information, and the dimension of the first state information accords with the dimension indicated by the first information. And the processing unit is used for determining a first model according to the first state information. And the transceiving unit is further used for sending third information to the second device, the third information comprises parameters of the first model, and the parameters of the first model comprise at least one of bias and weight.

In one possible implementation, the third information further includes information indicating a structure of the first model.

In one possible implementation, the information indicating the structure of the first model includes the number of layers of the first model, the type of neural network of each layer of the first model, the number of neurons, and the activation function. Alternatively, the information indicating the structure of the first model includes a model index.

In one possible implementation, the information for indicating the structure of the first model further includes jump connection indicating information, where the jump connection indicating information includes a start layer of the first model and an end layer of the first model, and the jump connection indicating information is used to indicate jump connection of the start layer and the end layer.

In one possible implementation, the second information further includes an action including an operation performed by the first device and a reward for describing an evaluation of the operation performed by the first device.

In a fourth aspect, there is provided a communication apparatus comprising: a processing unit and a transceiver unit.

And the receiving and transmitting unit is used for receiving first information from the first equipment, the first information indicates the dimension of first state information, and the first state information is input information of the first model. The first model includes a neural network for assisting the second device in making predictions or decisions. The receiving and transmitting unit is further configured to send second information to the first device, where the second information includes first state information, and a dimension of the first state information conforms to a dimension indicated by the first information. And the receiving and transmitting unit is also used for receiving the parameters of the first model from the first equipment. The parameters of the first model include at least one of bias and weight. And the processing unit is used for taking the first state information as the input information of the first model to obtain the output information of the first model.

In one possible implementation, the second information further includes an action including an operation performed by the first device and a reward for describing whether the operation performed by the first device was successful.

In one possible implementation, the second information is derived from a second model, which is trained by: the second state information is used as the input of the second model, and the output of the second model is the first state information. The dimension of the second state information is greater than the dimension of the first state information. And taking the first state information as the input of a third model, wherein the output of the third model is third state information, the dimension of the third state information is larger than that of the first state information, and the dimension of the third state information is equal to that of the second state information. The second model is trained by minimizing errors in the third state information and the second state information.

In a fifth aspect, a communication device is provided, which may be the communication device of any one of the third to fourth aspects in the above-described embodiments, or a chip provided in the communication device of any one of the third to fourth aspects. The communication device comprises a communication interface and a processor, and optionally a memory. The memory is used for storing a computer program or instructions or data, and the processor is coupled with the memory and the communication interface, when the processor reads the computer program or instructions or data, the communication device executes the method executed by the first device or the second device in the method embodiments of any one of the first aspect to the second aspect.

It will be appreciated that the communication interface may be implemented by an antenna, feeder, codec etc. in the communication device or, if the communication device is a chip provided in the first or second device, the communication interface may be an input/output interface of the chip, such as input/output pins etc. The communication means may further comprise a transceiver for the communication means to communicate with other devices. For example, when the communication apparatus is a first device, the other device is a second device. For another example, when the communication device is a second device, the other device is a first device.

In a sixth aspect, embodiments of the present application provide a chip system, where the chip system includes a processor and may further include a memory, to implement a method performed by a communication device in any one of the first to second aspects. In one possible implementation, the chip system further includes a memory for storing program instructions and/or data. The chip system may be formed of a chip or may include a chip and other discrete devices.

In a seventh aspect, the present application provides a computer readable storage medium storing a computer program or instructions which, when executed, implement the method performed by the first device in the above aspects; or implementing the method performed by the second device in the above aspects.

In an eighth aspect, there is provided a computer program product comprising: computer program code or instructions which, when executed, cause a method of the above aspects to be performed by a first device or cause a method of the above aspects to be performed by a second device to be performed.

In a ninth aspect, there is provided a communication device comprising units or modules for performing the methods of the above aspects.

In a tenth aspect, a communication device is provided that includes logic circuitry and an input-output interface. The logic unit is configured to perform a method performed by the first device or the second device in an embodiment of a method in any one of the first aspect to the second aspect. And the input/output interface is used for communicating with other devices through the communication device. For example, when the communication apparatus is a first device, the other device is a second device. For another example, when the communication device is a second device, the other device is a first device.

In an eleventh aspect, a communication system is provided. The communication system comprises a communication device according to any one of the possible implementation manners of the third aspect and a communication device according to any one of the possible implementation manners of the fourth aspect.

Advantageous effects of the above second to eleventh aspects and implementations thereof reference may be made to the description of the advantageous effects of the method of the first aspect and implementations thereof.

Drawings

FIG. 1 is a schematic diagram of a neural network architecture;

fig. 2 is a schematic diagram of a communication system according to an embodiment of the present application;

FIG. 3 is a schematic illustration of a scenario in which a central training distribution is performed;

fig. 4 is an exemplary flowchart of an information transmission method provided in an embodiment of the present application;

FIG. 5 is a schematic diagram of a first information provided in an embodiment of the present application;

FIG. 6 is a schematic diagram of a first information provided in an embodiment of the present application;

FIG. 7A is a schematic diagram of a neural network for determining second information according to an embodiment of the present application;

FIG. 7B is a schematic diagram of a self-encoder according to an embodiment of the present application;

fig. 8 is a schematic diagram of a second information provided in an embodiment of the present application;

FIG. 9 is a schematic diagram of a DQN provided in an embodiment of the present application;

fig. 10 is a schematic diagram of fifth information provided in an embodiment of the present application;

FIG. 11 is a schematic diagram of a third model according to an embodiment of the present disclosure;

FIG. 12 is a schematic diagram of a neural network for decision making or prediction by a second device according to an embodiment of the present application;

fig. 13 is a schematic diagram of a communication device according to an embodiment of the present application;

fig. 14 is a schematic diagram of a communication device according to an embodiment of the present application;

fig. 15 is a schematic diagram of a communication device according to an embodiment of the present application;

fig. 16 is a schematic diagram of a communication device according to an embodiment of the present application.

Detailed Description

Wireless communication has evolved rapidly, fifth generation mobile communication (5 ^th generation mobile communication technology, 5G) and sixth generation wireless fidelity (wireless fidelity, wi-Fi 6) standards have been commercially available, and next generation wireless technologies and standardization are under way worldwide. Wireless communication has penetrated various aspects of daily life and work as an integral part. With the rapid increase of the number of intelligent terminals and the popularization of internet of things (internet of things, ioT) devices, endless new wireless applications such as virtual reality, augmented reality and holograms are induced.

One task type of AI is prediction. Such as traffic prediction and channel prediction, etc. While another task type is decision making such as channel access, rate adaptation, power control and channel aggregation, etc. Existing AI algorithms can be divided into two categories, distributed training distribution execution and center training distribution execution. The central training service distributed execution means that the central node trains the neural network, and the parameters of the neural network are issued to the distributed nodes, and the distributed nodes conduct prediction or decision through the neural network. Distributed training distribution execution refers to training a neural network by distributed nodes and making predictions or decisions by the distributed nodes through the neural network. The central training profile execution may utilize global information, and thus may result in greater performance gains than the distributed training profile execution.

In order to facilitate understanding of the technical solutions provided in the embodiments of the present application, technical terms related to the embodiments of the present application are described below.

1) Models, referred to as AI models, may also be referred to as neural networks, are related to tasks performed by the device. For example, the device may perform the task of channel access, then the input information for the model may include carrier sense results, channel access behavior, and packet transmission results, and the output information may include a desired reward for accessing or not accessing the channel. For another example, the device may perform rate-adaptive tasks, and the input information for the model may include received power, carrier sense results, coded modulation scheme (modulation and coding scheme, MCS), and packet transmission results, and the output information may include probabilities of using the MCSs. For another example, the device may perform the task of channel aggregation, then the input information for the model may include the carrier sense results for each channel and the loading for each channel, and the output information may include the desired rewards using different channel aggregation methods. For another example, the device may perform the task of channel state information (channel state information, CSI) feedback, then the input information for the model may include a channel matrix and a beamforming matrix, and the output information may include compressed channel information. For another example, the device may perform the task of flow prediction, then the input of the model may include historical flow information and the output information may include predicted flow. The model may implement task related functions and may assist the device in making predictions or decisions.

Neural Networks (NNs) are a machine learning technique that simulates a human brain neural network in an effort to be able to implement artificial-like intelligence. The neural network may include an input layer, an intermediate layer (also referred to as a hidden layer), and an output layer. Taking the simplest neural network as an example, the internal structure and implementation will be described. Referring to fig. 1, fig. 1 is a schematic diagram of a fully connected neural network comprising 3 layers. As shown in fig. 1, the neural network includes 3 layers, namely an input layer, a hidden layer and an output layer, wherein the input layer has 3 neurons, the hidden layer has 4 neurons, the output layer has 2 neurons, and each layer of neurons is fully connected with the next layer of neurons. Each link between neurons corresponds to a weight that can be updated by training. Each neuron of the hidden layer and the output layer may also correspond to a bias, which may be updated by training. Updating the neural network refers to updating these weights and biases. In determining the structure of the neural network, i.e., the number of neurons included in each layer and how the outputs of the preceding neurons are input to the following neurons (i.e., the connection relationship between neurons), the entire information of the neural network can be determined with the parameters of the neural network, i.e., the weights and offsets, added.

2) The state information may be input information of the model, such as the aforementioned carrier sense result, channel access behavior, and packet transmission result. It is understood that the status information may be preset or predefined by the protocol.

3) The dimension of the state information refers to the number of neurons of the input layer of the model.

In addition, it should be understood that in the embodiments of the present application, at least one may also be described as one or more, and a plurality may be two, three, four or more, which is not limited in the present application. In the embodiment of the present application, "/" may indicate that the associated object is an "or" relationship, for example, a/B may indicate a or B; "and/or" may be used to describe that there are three relationships associated with an object, e.g., a and/or B, which may represent: there are three cases, a alone, a and B together, and B alone, wherein a, B may be singular or plural. In order to facilitate description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, the words "first", "second", etc. may be used to distinguish between technical features that are the same or similar in function. The terms "first," "second," and the like do not necessarily denote any order of quantity or order of execution, nor do the terms "first," "second," and the like. In this application embodiment, the terms "exemplary" or "such as" and the like are used to denote examples, illustrations, or descriptions, and any embodiment or design described as "exemplary" or "such as" should not be construed as preferred or advantageous over other embodiments or designs. The use of the word "exemplary" or "such as" is intended to present the relevant concepts in a concrete fashion to facilitate understanding.

In this application, "indication" may include direct indication, indirect indication, display indication, and implicit indication. When a certain indication information is described for indicating a, it can be understood that the indication information carries a, directly indicates a, or indirectly indicates a.

In the present application, information indicated by the indication information is referred to as information to be indicated. In a specific implementation process, there are various ways to indicate the information to be indicated, for example, but not limited to, the information to be indicated may be directly indicated, such as the information to be indicated itself or an index of the information to be indicated. The information to be indicated can also be indicated indirectly by indicating other information, wherein the other information and the information to be indicated have an association relation. It is also possible to indicate only a part of the information to be indicated, while other parts of the information to be indicated are known or agreed in advance. For example, the indication of the specific information may also be achieved by means of a pre-agreed (e.g., protocol-specified) arrangement sequence of the respective information, thereby reducing the indication overhead to some extent.

The information to be indicated can be sent together as a whole or can be divided into a plurality of pieces of sub-information to be sent separately, and the sending periods and/or sending occasions of the sub-information can be the same or different. The specific transmission method is not limited in this application. The transmission period and/or the transmission timing of the sub-information may be predefined, for example, predefined according to a protocol, or may be configured by the transmitting end device by transmitting configuration information to the receiving end device. The configuration information may include, for example, but not limited to, one or a combination of at least two of radio resource control signaling, medium access control (media access control, MAC) layer signaling, and physical layer signaling. Wherein radio resource control signaling such as packet radio resource control (radio resource control, RRC) signaling; the MAC layer signaling includes, for example, a MAC Control Element (CE); the physical layer signaling includes, for example, downlink control information (downlink control information, DCI).

Embodiments of the present application may be applicable in the context of a WLAN, for example, may be applicable in the institute of electrical and electronics engineers (Institute of Electrical and Electronics Engineers, IEEE) 802.11 system standard, such as the 802.11a/b/g, 802.11n, 802.11ac, 802.11ax standard, or its next generation, such as the 802.11be standard, wi-Fi 7 or very high throughput (extremely high throughput, EHT), 802.11ad,802.11ay,802.11bf, and further such as the 802.11be next generation, such as Wi-Fi 8 or ultra high reliability (Ultra High Reliability, UHR) or even next generation standards. Or the embodiment of the application can be also applied to wireless local area network systems such as an internet of things (internet of things, ioT) network or a Vehicle to X (V2X) network. Of course, the embodiments of the present application may also be applicable to other possible communication systems, such as LTE systems, LTE frequency division duplex (frequency division duplex, FDD) systems, LTE time division duplex (time division duplex, TDD), universal mobile telecommunication systems (universal mobile telecommunication system, UMTS), worldwide interoperability for microwave access (worldwide interoperability for microwave access, wiMAX) communication systems, 5G communication systems, future 6G communication systems, and the like.

The following takes a scenario in which the embodiments of the present application may be applied to WLAN as an example. It should be appreciated that WLANs start with the 802.11a/g standard, via 802.11n, 802.11ac, 802.11ax and 802.11be, which are now under discussion. Wherein 802.11n may also be referred to as High Throughput (HT); 802.11ac may also be referred to as very high throughput (very high throughput, VHT); 802.11ax may also be referred to as High Efficiency (HE) or Wi-Fi 6;802.11be may also be referred to as EHT or Wi-Fi 7, while standards prior to HT, such as 802.11a/b/g, etc., may be collectively referred to as Non-high throughput (Non-HT).

Referring to fig. 2, a network architecture diagram of a WLAN to which embodiments of the present application are applicable is shown. Fig. 2 illustrates an example in which the WLAN includes 1 wireless Access Point (AP) and 2 Stations (STAs). The STA associated with the AP can receive the radio frame transmitted by the AP and can also transmit the radio frame to the AP. In addition, the embodiments of the present application are equally applicable to communications between APs, for example, the APs may communicate with each other through a distributed system (distributed system, DS), and the embodiments of the present application are also applicable to communications between STAs. It should be understood that the number of APs and STAs in fig. 2 is by way of example only, and more or fewer may be provided.

The access point may be an access point for a terminal device (such as a mobile phone) to enter a wired (or wireless) network, and is mainly deployed in a home, a building and a park, where a typical coverage radius is several tens meters to hundreds meters, and of course, may also be deployed outdoors. The access point is equivalent to a bridge connecting a wired network and a wireless network, and is mainly used for connecting all wireless network clients together and then connecting the wireless network into an Ethernet. In particular, the access point may be a terminal device (e.g., a cell phone) or a network device (e.g., a router) with a Wi-Fi chip. The access point may be a device supporting the 802.11be standard. The access point may also be a device supporting multiple wireless local area network (wireless local area networks, WLAN) standards of 802.11 families, such as 802.11ax, 802.11ac, 802.11ad, 802.11ay, 802.11n, 802.11g, 802.11b, 802.11a, and 802.11be next generation. The access point in the present application may be a HE AP or an extremely high throughput (extremely high throughput, EHT) AP, or may be an access point that is adapted to future generations of Wi-Fi standards.

The station may be a wireless communication chip, a wireless sensor, a wireless communication terminal, or the like, and may also be referred to as a user. For example, the website may be a mobile phone supporting Wi-Fi communication function, a tablet computer supporting Wi-Fi communication function, a set top box supporting Wi-Fi communication function, a smart television supporting Wi-Fi communication function, a smart wearable device supporting Wi-Fi communication function, a vehicle communication device supporting Wi-Fi communication function, a computer supporting Wi-Fi communication function, and so on. Alternatively, the station may support 802.11be standard. Stations may also support multiple wireless local area network (wireless local area networks, WLAN) standards of 802.11 families, such as 802.11ax, 802.11ac, 802.11n, 802.11g, 802.11b, 802.11a, 802.11be next generation, etc.

The station in the present application may be an HE STA or an extremely high throughput (extremely high throughput, EHT) STA, and may also be an STA that is adapted to future generation Wi-Fi standards.

For example, the access points and sites may be devices applied in the internet of things, internet of things nodes, sensors, etc. in the internet of things (IoT, internet of things), smart cameras in smart homes, smart remote controls, smart water meter meters, sensors in smart cities, etc.

The AP and STA according to the embodiments of the present application may be an AP and STA applicable to the IEEE 802.11 system standard. An AP is a device deployed in a wireless communication network to provide wireless communication functions for its associated STA, and may be used as a backbone of the communication system, typically as a network side product supporting MAC and PHY of the 802.11 system standard, for example, may be a base station, a router, a gateway, a repeater, a communication server, a switch, or a bridge, where the base station may include various forms of macro base stations, micro base stations, relay stations, and so on. Here, for convenience of description, the above-mentioned devices are collectively referred to as an AP. STAs are typically end products supporting medium access control (media access control, MAC) and Physical (PHY) layers of the 802.11 system standard, such as cell phones, notebook computers, etc.

Referring to fig. 3, a Wi-Fi network is taken as an example, where at least one STA exists in the network, and each STA uses its own status information s _i To the central node (AP for example in fig. 3). The AP trains the neural network and trains the trained network parameters (NN as shown in fig. 3 _i ) Issuing to STA, then STA transmitting own state information s _i And inputting the neural network to obtain a decision result by reasoning. Where i=1, 2,3 ….

However, in the above solution, the status information of each STA, that is, the format of the input information of the neural network, which information the status information specifically includes, the dimension and the arrangement sequence of each status information, etc. need to be unified, which is inconvenient for the personalized algorithm design of each manufacturer. Moreover, each STA reports status information, and the AP needs more transmission resources and high overhead for transmitting relevant parameters of the neural network to the STA. For convenience in distinguishing the state information reported by the STA in the related art in the embodiments of the present application, the state information is referred to as original state information.

In view of this, an embodiment of the present application provides an information transmission method. In the method, the central node can indicate the dimension of the state information to be sent to the distributed node, and the distributed node can send the state information conforming to the dimension to the central node according to the indication of the central node. Based on the scheme, different distributed nodes can be allowed to use different state information, and meanwhile, the cost of reporting data can be effectively reduced because the dimension of the state information indicated by the central node is smaller than the dimension of the original state information sent by the distributed node to the central node in the related technology. In the method, the central node may determine a first model for performing a task based on the state information transmitted by the distributed node, and transmit the first model to the distributed node. Since the dimension of the state information of the first model is determined to be smaller than the dimension of the original state information used for training the model in the related art, the parameter quantity of the first model is also smaller than the parameter quantity of the model used for executing the task in the related art, and the cost for sending the model can be effectively reduced. In addition, the method does not need the distributed nodes to use the same state information, and can have a good privacy protection effect on the structure and parameters of the model.

Referring to fig. 4, an exemplary flowchart of an information transmission method according to an embodiment of the present application may include the following operations. In the embodiment shown in fig. 4, the central node may be a first device and the distributed nodes may be second devices. Wherein the first device may be an AP as shown in fig. 2 and the second device may be an STA as shown in fig. 2. Alternatively, the first device may be an STA as shown in fig. 2 and the second device may be an STA as shown in fig. 2.

S401: the first device sends first information to the second device.

Accordingly, the second device receives the first information from the first device.

Wherein the first information may indicate a dimension of the first state information. In one example, the first information may include a dimension index, which may correspond to a dimension of the state information. For example, the dimensions of the state information may be predefined, and each dimension of the state information may correspond to a dimension index. The first device may indicate the dimensions of the first state information to the second device via the dimension index. In another example, the first information may include indication information of a dimension of the first state information, which may indicate a value, such as 12, 13, 14, or 20. The value may be considered as a dimension of the first state information.

Alternatively, the dimension of the first state information may be smaller than the dimension of information reported by distributed nodes in the current technology of central training. Taking the rate adaptive scenario as an example, assuming that the original state information that needs to be reported by the distributed nodes in the rate adaptive scenario is 240 dimensions, in the case of the rate adaptive scenario in the embodiment shown in fig. 4, the dimension of the first state information may be smaller than 240 dimensions.

In a possible scenario, the first device may carry the first information with a newly added information element. For example, a new information element may be predefined by the protocol, which information element is used to carry the first information. Referring to fig. 5, an exemplary format of an information element is shown. As shown in fig. 5, the information element may include an element ID (element ID), a length (length), an element ID extension (extension), and first information. The first information may include a dimension of the state information, for indicating the dimension of the first state information. It will be appreciated that the format of the information element shown in fig. 5 is shown by way of example only and is not limiting of the information element.

In another possible scenario, the first device may employ an existing information element to carry the first information. For example, the first device may carry the first information using an HT control (control) field (field). Referring to fig. 6, the first device may use an a-control subfield (a-control subfield) in the HT control field to carry the first information. As shown in fig. 6, the first device may carry first information in a control information (control information) field indicating a dimension of the first state information.

Optionally, in the embodiment shown in fig. 4, the following operation S400 may also be included.

S400: the second device transmits a first frame to the first device.

Accordingly, the first device receives the first frame from the second device.

The first frame may be used to indicate a task performed by the second device, or the first frame may be used to indicate that the second device requests an AI-assisted function/task to be turned on. For example, the first frame may be a function request frame requesting an open AI-assisted function/task. The first device may determine, based on the first frame, an AI-assisted function/task that the second device requests to be turned on, such as AI-assisted channel access, AI-assisted rate adaptation, AI-assisted channel aggregation, AI-assisted CSI feedback, or AI-assisted traffic prediction, etc. Alternatively, in the above, the functions/tasks of the AI-assistance request by the second device may be predefined functions/tasks, and the first frame may carry a function identifier for indicating that the second device requests the AI-assistance request to be turned on.

Optionally, the first device may determine the first model according to the AI-assisted function/task requested to be turned on by the second device, where the first model may implement the AI-assisted function/task requested to be turned on by the second device, and assist the second device in making decisions or predicting, that is, assist the second device to perform a corresponding task. It is understood that the first model may include a neural network to be trained. The first state information in S401 may be for training the neural network to be trained.

Based on operation S400, the first device may determine a task performed by the second device, so that the first device may determine a function for which the second device requests an open AI-assistance, so that the first device may determine a first model for implementing the function.

S402: the second device sends second information to the first device.

Accordingly, the first device receives the second information from the second device.

The second information may include first status information. It is understood that the dimension of the first state information contained in the second information may correspond to the dimension indicated by the first information. For example, in the case where the dimension indicated by the first information is 20 dimensions, the dimension of the first state information included in the second information is 20 dimensions. For another example, in the case where the dimension indicated by the first information is 40 dimensions, the dimension of the first state information included in the second information is 40 dimensions.

Optionally, the second information may further include an action and a reward. Wherein the action may include an operation performed by the second device, rewarding an evaluation describing the operation performed by the first device. For example, taking a channel access scenario as an example, actions may include accessing a channel and not accessing a channel. The reward is 1 in the case of an access channel and successful packet transmission, is-1 in the case of an access channel and failed packet transmission, and is 0 in the case of no access channel. As another example, taking a rate adaptation scenario as an example, the action may include which MCS to use and the reward may include the rate at which to use. As another example, taking a channel aggregation scenario as an example, an action may include which channel aggregation scheme to use, and a reward may include a rate at which the channel aggregation scheme is used. In the above, the actions and rewards may be used to train the first model using a reinforcement learning method.

Optionally, the second information may further include a tag. The tag is used to indicate the desired output information of the first model. The tag may be used to train the first model using a supervised learning approach. For example, taking a traffic prediction scenario as an example, the tag may be current traffic information and/or traffic information for a future period of time. The input information of the first model may be flow information for a period of time historic and the output information may be flow information predicted by the first model. The first device may train the first model by minimizing errors between the output information and the labels, such as mean square error (mean square error, MSE).

It should be noted that the second information may be reported in real time, or may be reported in batch. The real-time reporting may be understood as sending the second information to the first device every time the second device makes a decision. For example, taking a channel access scenario as an example, the second device may send the second information to the first device after deciding to access the channel or deciding not to access the channel. Batch reporting may be understood as that the second device sends the second information after multiple decisions at one time, e.g. the second device may periodically send the second information to the first device.

In a possible implementation manner, the second device may process the first state information in the related art, so as to obtain the first state information conforming to the dimension indicated by the first information. For example, the second device may perform compression processing on the first state information, compressing the dimension of the first state information to the dimension indicated by the first information.

The method of processing the first state information by the second device will be described below by taking an auto encoder (auto encoder) as an example. It may be appreciated that the second device may also obtain the first state information corresponding to the dimension indicated by the first information in other manners, such as dimension reduction, compression, and the like, which are not limited in this application.

The following describes a manner of processing the first state information by the second device with reference to fig. 7A and 7B. The first device may perform the dimension reduction processing on the first state information through the neural network illustrated in fig. 7A. Fig. 7A shows a neural network in which the number of neurons of the input layer is 240, that is, the dimension of the input first state information is 240 dimensions. In the embodiment shown in fig. 7A, the dimension indicated by the first information is exemplified as 20 dimensions. The dimension of the first state information can be reduced to 128 dimensions through the first hiding layer, the dimension of the first state information can be reduced to 64 dimensions through the second hiding layer, and the dimension of the first state information can be reduced to 20 dimensions conforming to the first information indication through the third hiding layer. Through the neural network shown in fig. 7A, the 240-dimensional first state information can be reduced in size to 20-dimensional first state information. The second device may send 20-dimensional first state information to the first device.

The neural network shown in fig. 7A may be part of a first device trained self-encoder. A training method of the self-encoder is shown below by fig. 7B. As shown in fig. 7B, for convenience of description, the self-encoder may be divided into two models, a portion of an encoder (encoder) (s→h) may be referred to as a second model (self-encoder shown in fig. 7A), and a decoder (decoder) (h→s') may be referred to as a third model.

In fig. 7B, the second model may be implemented as a dimension reduction operation for the first state information. The input information of the second model may be raw state information (240-dimensional is an example in fig. 7B). After the second model, second state information after the dimension reduction can be obtained (for example, 20 dimensions in fig. 7B). The second device may take the output information of the second model as input information of the third model. The third model may implement an operation of upscaling the second state information. The second state information subjected to the second model dimension reduction is subjected to dimension increasing, and third state information (240 dimensions are taken as an example in fig. 7B) with the same dimension as the input information of the second model is obtained after the third model. In other words, the third model attempts to restore the second state information after the dimension reduction to the original state information before the dimension reduction.

The second device may train the second model by minimizing an error between input information of the second model, such as the second state information, and output information of the third model, such as the third state information. For example, the second device may minimize the loss function using a random gradient descent method, i.e., minimize the mean square error between S' and S, i.e., min _θ (S′-S) ² The neural network parameter θ is updated to train the second model.

It will be appreciated that the second device may train the third model using a variety of learning methods, such as supervised learning, unsupervised learning, or reinforcement learning, with random gradient descent methods to minimize the loss function to train the second model, as the application is not specifically limited. Similarly, the second device can customize the input first state information and the customized neural network structure, and only needs to obtain the first state information which accords with the dimension indicated by the first information.

In a possible implementation, the second device may employ a newly added information element, such as the information element shown in fig. 5, to carry the second information. In another possible implementation, the second device may use an existing field to carry the second information, as shown in fig. 6, and an a-control subfield (a-control subfield) in the HT control field to carry the second information.

In one example, the second device may send a set of AI functions (tasks) to the first device. The AI functionality (task) set may include one or more AI functionalities/tasks. The set of AI functions (tasks) may be used to indicate that the first state information is for one or more AI functions/tasks. That is, the second device may indicate to the first device that the transmitted first state information is used to train a model of the AI function/task. In another example, the second device may send the first device the amount of the first state information and/or the dimension of the first state information.

Referring to fig. 8, the second device may transmit the first status information to the first device through the format shown in fig. 8. In the embodiment illustrated in fig. 8, the AI task set field may indicate that the first state information is to be used to train a model of AI functions/tasks contained by the AI task set. The data number field is used to indicate the number N of first state information contained in the frame, and the data dimension field may be used to indicate the dimension of the first state information. m is m _i (i=1, 2,3, …, N) represents first state information.

S403: the first device determines a first model based on the first state information.

In the above, the first device may determine the first model according to the AI-assisted function/task that the first device requests to be turned on. In S403, the first device may train the first model by using the first state information as input information of the first model according to the first state information reported by the second device in S402. The first device may stop training the first model when the loss function of the first model is less than or equal to the first threshold, or when the first model adopts a training method of gradient descent, and the number of iterations of gradient descent is greater than or equal to the second threshold.

Optionally, the first device may also train the first model based on the actions and rewards. For example, the first model may train the first model using a reinforcement learning method, and the first device may determine a loss function of the first model based on the actions and rewards, which may be used to train the first model using the reinforcement training method. Without loss of generality, most reinforcement learning methods are trained based on state s, action a, rewards r. The training method of the first model will be described below by taking deep reinforcement learning (deep reinforcement learning, DQN) as an example.

Referring to FIG. 9, a schematic diagram of the working principle of DQN is shown, including a target Q network and a predicted Q network, both of which have network structuresThe initial parameters are the same, but the predicted Q network updates the parameters (denoted as θ) for each training, while the target Q network parameters (denoted as θ ^- ) The update is performed after every C training runs, typically c=100. It is understood that C may be set according to an empirical value, or may be set to an integer of 50, 200, or the like. In the above, the predictive Q network may be the first model to be trained in embodiments of the present application. The target Q network may be used to assist in training the predictive Q network.

The training data used by DQN can be referred to as experience (experience), i.e. e _t ＝(s _t ,a _t ,r _t ,s _t+1 ). In this embodiment of the present application, the experience may be the second information sent by the second device to the first device, for example, or may be generated according to the second information. The first device stores the experience in an experience pool for training of the neural network. The training method may use a small batch gradient descent method (mini-batch gradient descent) to update neural network parameters (e.g., weights and biases) by minimizing the loss function. The neural network has a loss function ofWhere γ represents a discount factor, typically γ=0.9, b represents batch data (e.g., second information) randomly sampled from the experience pool, r _t Rewards obtained for current actions, e.g. rewards contained in the second information, Q (s _t ,a _t The method comprises the steps of carrying out a first treatment on the surface of the θ) is the output of the predictive Q network, max _a′ Q(s _t+1 ,a′；θ ^- ) Is the output of the target Q network.

S404: the first device sends third information to the second device.

Accordingly, the second device receives the third information from the first device.

In the above, the third information may include parameters of the first model. For example, the third information may include at least one of a weight and a bias. For example, the third information may indicate weights of the neurons of the first model and connections between the neurons. For another example, the third information may indicate a bias of neurons of the first model.

In a possible implementation, the first device may also send information indicating the structure of the first model to the second device. For convenience of description, information indicating the structure of the first model will be hereinafter referred to as fifth information. For example, the first device may carry the fifth information in the third information. For another example, the first device may transmit the fifth information to the second device after performing S401 and before performing S403. For another example, the first device may transmit the fifth information to the second device after performing S403 and before performing S404. For another example, the first device may transmit fifth information to the second device after performing S404.

In a possible implementation, the first device may use a newly added information element, such as the information element shown in fig. 5, to carry the fifth information. In another possible implementation, the first device may use an existing field to carry the fifth information, as shown in fig. 6, and an a-control subfield (a-control subfield) in the HT control field to carry the fifth information.

In one example, the fifth information may include a model index. For example, models of various structures may be predefined or preconfigured by a protocol, each model corresponding to a model index. Thus, the first device may indicate the structure of the first model to the second device through the model index. The second device may determine the first model by parameters of the first model and a structure of the first model.

In another example, the fifth information may include a number of layers of the first model, a neural network type of each layer of the first model, a number of neurons of each layer of the first model, and an activation function of each layer of the first model. Since the neural network may include a plurality of layers, if the structure of one neural network is to be determined, the structure of each layer needs to be determined. If it is desired to determine the structure of a layer of neural network, it is necessary to determine the type of neural network, the number of neurons, the activation function, etc. used for the layer. Thus, the structure of the first model can be determined by the third information.

Alternatively, if a layer in the first model is a convolutional neural network, the fifth information may further include a size and a step size of a convolutional kernel of the convolutional neural network. Alternatively, for a neural network of the same structure of successive layers, the layer type and the number of layers may be included in the third information.

Referring to fig. 10, an exemplary diagram of fifth information in an embodiment of the present application is shown. As shown in fig. 10, the fifth information may include a type number, that is, the structure of the first model indicated by the information includes several types of neural networks. For example, layer types may include fully connected layers (fully connected layer, FC), convolutional layers (convolutional layer), gated loop units (gated recurrent unit, GRU), long and short time memories (long short term memory, LSTM), and the like.

The fifth information further includes structure information corresponding to each type. Taking type 1 as an example, type 1 is a convolution layer, then the structural information corresponding to the convolution layer may include the number of layers, i.e., the convolution layer is the first layer of the first model, the number of neurons, and the activation function. Since type 1 is a convolution layer, the structure information may also contain information of the size and step size of the convolution kernel.

In one possible implementation, the fifth information may further include skip connection (skip connection) indication information. Jump connection is a common structure in convolutional networks and fully-connected networks, which is beneficial to the fusion of bottom-layer features and high-layer features, and can improve the learning ability of the neural network. In the case that the layer of the jump connection is included in the first model, jump connection instruction information may be added to the third information. For example, the jump connection indication information may contain a start layer of the jump connection and an end layer of the jump connection, that is to say the start layer and the end layer are jump connected. Optionally, the hop connection indication information may further include a hop connection number for indicating that several hop connections exist in the first model.

Optionally, as shown in fig. 10, the fifth information may include fourth information, where the fourth information may be used to indicate whether the first model includes a hopping connection, or indicate whether the fifth information includes hopping connection indication information. If the fourth information indicates that the fifth information does not contain the jump connection indication information, the first model can be considered to contain no jump connection, and if the jump connection indication information does not exist in the fifth information, the cost can be saved, and the processing resource of the second device can be saved. If the fourth information indicates that the information contains the hopping connection indication information, then the first model can be considered to contain no hopping connection, and the hopping connection indication information is present in the fifth information (as shown in fig. 10).

For example, the fourth information may be implemented by 1-bit (bit) information, where the fourth information may indicate that the information does not include the jump connection indication information when the value of the fourth information is "0", and the fourth information may indicate that the fifth information includes the jump connection indication information when the value of the fourth information is "1". Otherwise, if the fourth information is "1", the fifth information may be indicated as not including the jump connection indication information, and if the fourth information is "0", the fifth information may be indicated as including the jump connection indication information.

In a possible scenario, the first device may carry fifth information with the newly added information element, indicating to the second device the structure of the first model. For example, the newly added information element may include an element ID, a length, an element ID extension, and fifth information, as shown in fig. 5. In another possible scenario, the second device may use an existing field to carry fifth information, as shown in fig. 6, an a-control subfield (a-control subfield) in the HT control field to carry fifth information.

Optionally, the embodiment shown in fig. 4 may further include the following operation S405.

S405: and the second equipment makes decisions or predicts according to the first model.

For example, the second device may input the first state information to the first model, resulting in output information of the first model. The output information of the first model may assist the second device in making decisions or predictions.

Taking the rate adaptation scenario as an example, the second device may input MCS, ACK, received power, contention waiting period, carrier sense result, and carrier sense duration as first state information into the first model as shown in fig. 11. The first model is illustrated in fig. 11 as comprising two hidden layers. The first model shown in fig. 11 may be trained by a first device and sent to a second device. The first model may output a probability of transmission using each MCS or a desired prize transmitted using each MCS based on the first status information. The second device may make a decision based on the output of the first model. For example, the second device may select the MCS with the highest probability or highest desired prize among the outputs of the first model for transmission.

In a possible scenario, in order to reduce the error of the output information of the first model and increase the accuracy of the decision or prediction of the first model, the second device may combine the second model with the first model to make the decision or prediction. Referring to fig. 12, the second device may input the first state information before the dimension reduction to the second model, and use the output information of the second model as the input information of the first model. For example, in fig. 12, the input state information S is [ MCS, ACK, received power, contention waiting period, (carrier sense result, carrier sense duration) ×10] ×10, and the total is 240 dimensions, and after the second model performs dimension reduction, the output h of the second model is 20 dimensions. The second device takes the output h of the second model as input information for the first model to obtain the desired rewards/probabilities for transmission using the respective MCSs.

Based on the scheme, different distributed nodes can be allowed to use different state information, and meanwhile, the dimension of the state information indicated by the central node is smaller than that of the state information sent by the distributed node to the central node in the related art, so that the quantity of the state information of the distributed node in the scheme is also smaller than that of the state information sent by the distributed node to the central node in the related art, and the overhead of reporting data is effectively reduced. In addition, since the number of the state information of the first model is determined to be smaller than the number of the state information for training the model in the related art, the parameter amount of the first model is also smaller than the parameter amount of the model for performing the task in the related art, and the overhead of transmitting the model can be effectively reduced. In addition, the scheme does not need the distributed nodes to use the same state information, and can have a good privacy protection effect on the structure and parameters of the model.

Communication devices for implementing the above method in the embodiments of the present application are described below with reference to the accompanying drawings. Therefore, the above contents can be used in the following embodiments, and repeated contents are not repeated.

Fig. 13 is a schematic block diagram of a communication device 1300 provided in an embodiment of the present application. The communications apparatus 1300 may correspond to implementing the functions or steps implemented by the first device or the second device in the above-described method embodiments. The communication device may include a processing unit 1310 and a transceiving unit 1320. Optionally, a storage unit may be included, which may be used to store instructions (code or programs) and/or data. The processing unit 1310 and the transceiving unit 1320 may be coupled to the storage unit, for example, the processing unit 1310 may read instructions (codes or programs) and/or data in the storage unit to implement the corresponding method. The units can be independently arranged or partially or fully integrated.

In some possible embodiments, the communications apparatus 1300 can correspondingly implement the behavior and functions of the first device in the method embodiments described above. For example, the communication apparatus 1300 may be a first device, or may be a component (e.g., a chip or a circuit) applied to the first device. The transceiving unit 1320 may be used to perform all the receiving or transmitting operations performed by the first device in the embodiment shown in fig. 4. Such as SS401 or S402 or S404 in the embodiment shown in fig. 4, and/or other processes for supporting the techniques described herein; wherein the processing unit 1310 is configured to perform all operations performed by the first device in the embodiment shown in fig. 4, except for the transceiving operations, such as S403 in the embodiment shown in fig. 4, and/or other procedures for supporting the techniques described herein.

The transceiver 1320 is configured to send first information to the second device, where the first information indicates a dimension of first state information, and the first state information is input information of the first model. The first model includes a neural network for assisting the second device in making predictions or decisions. The transceiver 1320 is further configured to receive second information from the second device, where the second information includes first status information, and a dimension of the first status information matches a dimension indicated by the first information. The processing unit 1310 is configured to determine a first model according to the first state information. The transceiver unit 1320 is further configured to send third information to the second device, where the third information includes parameters of the first model, and the parameters of the first model include at least one of bias and weight.

In some possible embodiments, the communications apparatus 1300 can correspondingly implement the behavior and functions of the second device in the method embodiments described above. For example, the communication apparatus 1300 may be a second device, or may be a component (e.g., a chip or a circuit) applied to the second device. The transceiving unit 1320 may be used to perform all the receiving or transmitting operations performed by the second device in the embodiment shown in fig. 4. Such as S401 or S402 or S404 in the embodiment shown in fig. 4, and/or other processes for supporting the techniques described herein; wherein the processing unit 1310 is configured to perform all operations performed by the second device except for the transceiving operations, such as S405 in the embodiment shown in fig. 4, and/or other procedures for supporting the techniques described herein.

The transceiver 1320 is configured to receive first information from a first device, where the first information indicates a dimension of first state information, and the first state information is input information of a first model. The first model includes a neural network for assisting the second device in making predictions or decisions. The transceiver 1320 is further configured to send second information to the first device, where the second information includes first state information, and a dimension of the first state information conforms to a dimension indicated by the first information. The transceiver unit 1320 is further configured to receive parameters of the first model from the first device. The parameters of the first model include at least one of bias and weight. The processing unit 1310 is configured to obtain output information of the first model by using the first state information as input information of the first model.

For the operations performed by the processing unit 1310 and the transceiving unit 1320, reference may be made to the relevant description of the method embodiments described previously.

It should be appreciated that the processing unit 1310 in the embodiments of the present application may be implemented by a processor or a processor-related circuit component, and the transceiver unit 1320 may be implemented by a transceiver or a transceiver-related circuit component or a communication interface.

Based on the same concept, as shown in fig. 14, the present embodiment provides a communication apparatus 1400. The communication device 1400 includes a processor 1410. Optionally, the communications device 1400 may also include a memory 1420 for storing instructions to be executed by the processor 1410 or for storing input data required by the processor 1410 to execute instructions or for storing data generated after the processor 1410 executes instructions. The processor 1410 may implement the method shown in the method embodiment described above through instructions stored in the memory 1420.

Based on the same concept, as shown in fig. 15, the present embodiment provides a communication device 1500, and the communication device 1500 may be a chip or a chip system. Alternatively, the chip system in the embodiments of the present application may be formed by a chip, and may also include a chip and other discrete devices.

The communications device 1500 may include at least one processor 1510, the processor 1510 being coupled to a memory, which may optionally be located within the device or external to the device. For example, the communications apparatus 1500 can also include at least one memory 1520. Memory 1520 holds computer programs, configuration information, computer programs or instructions and/or data necessary to implement any of the embodiments described above; the processor 1510 may execute a computer program stored in the memory 1520 to perform the method of any one of the embodiments described above.

The coupling in the embodiments of the present application is an indirect coupling or communication connection between devices, units, or modules, which may be in electrical, mechanical, or other forms for information interaction between the devices, units, or modules. The processor 1510 may operate in conjunction with the memory 1520. The specific connection medium between the transceiver 1530, the processor 1510, and the memory 1520 is not limited in this embodiment.

The communication apparatus 1500 may further include a transceiver 1530, and the communication apparatus 1500 may perform information interaction with other devices through the transceiver 1530. The transceiver 1530 may be a circuit, bus, transceiver, or any other means that may be used for information interaction, or referred to as a signal transceiver unit. As shown in fig. 15, the transceiver 1530 includes a transmitter 1531, a receiver 1532, and an antenna 1533. In addition, when the communication device 1500 is a chip-type device or circuit, the transceiver in the communication device 1500 may be an input/output circuit and/or a communication interface, and may input data (or receive data) and output data (or transmit data), and the processor may be an integrated processor or a microprocessor or an integrated circuit, and the processor may determine the output data according to the input data.

In a possible implementation manner, the communication apparatus 1500 may be applied to the first device, and in particular, the communication apparatus 1500 may be the first device, or may be an apparatus capable of supporting the first device to implement the function of the first device in any of the foregoing embodiments. Memory 1520 holds the necessary computer programs, computer programs or instructions and/or data to implement the functions of the first device in any of the embodiments described above. The processor 1510 may execute a computer program stored in the memory 1520 to perform the method performed by the first device in any of the embodiments described above.

In another possible implementation manner, the communication apparatus 1500 may be applied to the second device, and in particular, the communication apparatus 1500 may be the second device, or may be an apparatus capable of supporting the second device, and implementing the function of the second device in any of the foregoing embodiments. Memory 1520 holds the necessary computer programs, computer programs or instructions and/or data to implement the functions of the second device in any of the embodiments described above. The processor 1510 may execute a computer program stored in the memory 1520 to perform the method performed by the second device in any of the embodiments described above.

Since the communication apparatus 1500 provided in this embodiment may be applied to a first device, a method performed by the first device is completed, or applied to a second device, a method performed by the second device is completed. Therefore, reference may be made to the above method embodiments for the technical effects, which are not described herein.

In the embodiments of the present application, the processor may be a general purpose processor, a digital signal processor, an application specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and may implement or execute the methods, steps, and logic blocks disclosed in the embodiments of the present application. The general purpose processor may be a microprocessor or any conventional processor or the like. The steps of a method disclosed in connection with the embodiments of the present application may be embodied directly in a hardware processor for execution, or in a combination of hardware and software modules in the processor for execution.

In the embodiment of the present application, the memory may be a nonvolatile memory, such as a hard disk (HDD) or a Solid State Drive (SSD), or may be a volatile memory (volatile memory), for example, a random-access memory (RAM). The memory may also be any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such. The memory in the embodiments of the present application may also be circuitry or any other device capable of implementing a storage function, for storing a computer program, a computer program or instructions and/or data.

Based on the above embodiments, referring to fig. 16, another communication device 1600 is also provided in the embodiments of the present application, including: an input-output interface 1610 and logic circuitry 1620; an input/output interface 1610, configured to receive a code instruction and transmit the code instruction to the logic circuit 1620; logic circuitry 1620 to execute code instructions to perform the methods performed by the first device or the second device in any of the embodiments described above.

Hereinafter, an operation performed by the communication apparatus applied to the first device or the second device will be described in detail.

In an alternative embodiment, the communication apparatus 1600 may be applied to a first device, and perform a method performed by the first device, for example, a method performed by the first device in the embodiment shown in fig. 4.

An input-output interface 1610, configured to send first information to the second device, where the first information indicates a dimension of first state information, and the first state information is input information of the first model. The first model includes a neural network for assisting the second device in making predictions or decisions. The input/output interface 1610 is further configured to receive second information from the second device, where the second information includes first state information, and a dimension of the first state information conforms to a dimension indicated by the first information. Logic circuitry 1620 is configured to determine a first model based on the first state information. The input-output interface 1610 is further configured to send third information to the second device, where the third information includes parameters of the first model, and the parameters of the first model include at least one of a bias and a weight.

In another alternative embodiment, the communication apparatus 1600 may be applied to a second device, and perform a method performed by the second device, for example, a method performed by the second device in the foregoing method embodiment shown in fig. 4.

The input-output interface 1610 is configured to receive first information from a first device, where the first information indicates a dimension of first state information, and the first state information is input information of a first model. The first model includes a neural network for assisting the second device in making predictions or decisions. The input/output interface 1610 is further configured to send second information to the first device, where the second information includes first state information, and a dimension of the first state information conforms to a dimension indicated by the first information. The input-output interface 1610 is further configured to receive parameters of the first model from the first device. The parameters of the first model include at least one of bias and weight. The logic circuit 1620 is configured to obtain output information of the first model by using the first state information as input information of the first model.

Since the communication apparatus 1600 provided in this embodiment may be applied to a first device, a method performed by the first device is completed, or applied to a second device, a method performed by the second device is completed. Therefore, reference may be made to the above method embodiments for the technical effects, which are not described herein.

Based on the above embodiments, the embodiments of the present application further provide a communication system. The communication system comprises at least one communication means applied to a first device and at least one communication means applied to a second device. The technical effects obtained can be referred to the above method embodiments, and will not be described herein.

Based on the above embodiments, the present application further provides a computer readable storage medium storing a computer program or instructions that, when executed, cause a method performed by a first device or a method performed by a second device in any of the above embodiments to be performed. The computer readable storage medium may include: various media capable of storing program codes, such as a U disk, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk.

In order to implement the functions of the communication device of fig. 13 to 16, the embodiment of the application further provides a chip, which includes a processor, and is configured to support the communication device to implement the functions related to the first device or the second device in the method embodiment. In one possible design, the chip is connected to a memory or the chip comprises a memory for holding the necessary computer programs or instructions and data for the communication device.

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer programs or instructions. These computer programs or instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer programs or instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer programs or instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

It will be apparent to those skilled in the art that various modifications and variations can be made to the embodiments of the present application without departing from the scope of the embodiments of the present application. Thus, if such modifications and variations of the embodiments of the present application fall within the scope of the claims and the equivalents thereof, the present application is intended to encompass such modifications and variations.

Claims

1. An information transmission method, comprising:

the method comprises the steps that a first device sends first information to a second device, wherein the first information indicates the dimension of first state information, and the first state information is input information of a first model; the first model includes a neural network for assisting the second device in making predictions or decisions;

the first device receives second information from the second device, the second information comprising the first state information, the dimension of the first state information conforming to the dimension indicated by the first information;

the first device determines a first model according to the first state information;

the first device sends third information to the second device, the third information including parameters of the first model including at least one of bias and weight.

2. The method of claim 1, wherein the third information further comprises information indicating a structure of the first model.

3. The method of claim 2, wherein the information indicative of the structure of the first model includes a number of layers of the first model, a neural network type of each layer of the first model, a number of neurons, and an activation function; or alternatively

The information indicating the structure of the first model includes a model index.

4. The method of claim 3, wherein the information indicating the structure of the first model further comprises a jump connection indication information including a start layer of the first model and an end layer of the first model, the jump connection indication information indicating jump connection of the start layer and the end layer.

5. The method of any of claims 1-4, wherein the second information further comprises an action including an operation performed by the first device and a reward for describing an evaluation of the operation performed by the first device.

6. The method according to any one of claims 1 to 5, wherein the second information further includes task information, and the task information is used to indicate a task corresponding to the first state information.

7. An information transmission method, comprising:

the second device receives first information from the first device, the first information indicating a dimension of first state information, the first state information being input information of a first model; the first model includes a neural network for assisting the second device in making predictions or decisions;

The second device sends second information to the first device, the second information comprises the first state information, and the dimension of the first state information accords with the dimension indicated by the first information;

the second device receives parameters of a first model from the first device; the parameters of the first model include at least one of bias and weight;

and the second equipment inputs the first state information into the first model to obtain output information of the first model, wherein the output information is used for the decision or prediction of the second equipment.

8. The method of claim 7, wherein the third information further comprises information indicating a structure of the first model.

9. The method of claim 8, wherein the information indicative of the structure of the first model includes a number of layers of the first model, a neural network type of each layer of the first model, a number of neurons, and an activation function; or alternatively

10. The method of claim 9, wherein the information indicating the structure of the first model further comprises a jump connection indication information including a start layer of the first model and an end layer of the first model, the jump connection indication information indicating jump connection of the start layer and the end layer.

11. The method of any of claims 7-10, wherein the second information further comprises an action including an operation performed by the first device and a reward for describing an evaluation of the operation performed by the first device.

12. The method according to any one of claims 7 to 11, wherein the second information further includes task information, and the task information is used to indicate a task corresponding to the first state information.

13. The method according to any one of claims 7 to 12, wherein the second information is derived from a second model, the second model being trained by:

taking second state information as input of the second model, wherein output of the second model is the first state information; the dimension of the second state information is larger than the dimension of the first state information;

taking the first state information as input of a third model, wherein the output of the third model is third state information, the dimension of the third state information is larger than that of the first state information, and the dimension of the third state information is equal to that of the second state information;

The second model is trained by minimizing errors of the third state information and the second state information.

14. A communication device comprising means for performing the method of any one of claims 1-6 or means for performing the method of any one of claims 7-13.

15. A communication device, comprising: a processor and a memory;

the memory is used for storing a computer program or instructions;

the processor being configured to execute a computer program or instructions in a memory to cause the apparatus to perform the method of any one of claims 1 to 6 or to cause the apparatus to perform the method of any one of claims 7 to 13.

16. A computer readable storage medium storing computer executable instructions which, when invoked by an electronic device, cause the electronic device to perform the method of any one of claims 1 to 6 or the electronic device to perform the method of any one of claims 7 to 13.

17. A computer program product comprising computer-executable instructions which, when run on a computer, cause the computer to perform the method of any one of claims 1 to 6 or cause the computer to perform the method of any one of claims 7 to 13.

18. A communication system comprising means for performing the method of any of claims 1-6 and comprising means for performing the method of any of claims 7-13.