Task-based collaborative reasoning methods, devices, network equipment, storage media, and program products
By dynamically allocating AI tasks in a collaborative inference mode between the terminal and edge servers and devices, the limitations of terminal computing power and communication resources are solved, achieving reasonable allocation of AI tasks and efficient computation offloading, thus improving system performance.
Patent Information
- Application Number
- CN202510631993.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-16
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, collaborative inference methods between terminals and edge servers and between terminals are limited by communication spectrum resources and terminal computing power, making it difficult to efficiently orchestrate and offload AI tasks, and failing to guarantee computing power and response speed.
By opening up the AI capabilities of user devices and network devices and the first task strategy to the first application, and adopting a three-way collaborative inference mode of terminal and edge server, device-to-device and terminal-terminal-base station, computing tasks are dynamically allocated to achieve model segmentation and computing offloading, thus avoiding overload of individual devices.
It enables the rational allocation of AI tasks under the constraints of communication and computing resources, avoiding device overload and improving the overall system's computing power and response speed.
Smart Images

Figure CN123124014A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of wireless technology, and in particular to a task collaborative reasoning method, apparatus, network device, storage medium, and program product. Background Technology
[0002] In AI-driven applications, particularly in image recognition and natural language processing, the increasing complexity of deep neural network (DNN) inference tasks highlights the limitations of single-device computing power. Currently, employing collaborative inference to distribute computational load between the terminal and edge computing, thereby balancing the computational load and improving inference performance, has become a key approach to addressing these challenges. A crucial technology based on this approach is AI model segmentation, which divides the inference task into multiple parts for computation on different devices.
[0003] Existing collaborative inference methods include terminal-to-edge server collaborative inference and terminal-to-terminal collaborative inference. However, the terminal-to-edge server inference method is affected by communication spectrum resources and channel conditions, thus limiting the transmission rate. The terminal-to-terminal inference method is affected by the limited computing power of individual terminal devices. Neither method can guarantee efficient orchestration and computation offloading of AI tasks. Therefore, it is necessary to improve the collaborative inference method for AI tasks to compute effective computation offloading strategies, rationally allocate computing tasks to avoid overloading of individual devices, and improve the overall system's computing power and response speed. Summary of the Invention
[0004] The purpose of this application is to provide a task collaborative reasoning method, apparatus, network device, storage medium, and program product, which can achieve reasonable allocation of AI tasks under the constraints of communication resources and computing resources, so as to avoid the problem of overload of a single device.
[0005] One embodiment of this application provides a task collaborative reasoning method, wherein the method is executed by a first device and includes:
[0006] Sending first information to a first application; wherein the first information is used by the first application to determine a second strategy for a first task, and the first information includes one or more of the following:
[0007] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0008] AI capability information for at least one network device;
[0009] The first strategy for the first task.
[0010] Optionally, the task collaborative reasoning method further includes:
[0011] The first strategy is determined based on one or more of the AI capability information of the at least one user device, the AI capability information of the at least one network device, and the requirements of the first task.
[0012] Optionally, in the task collaborative reasoning method, the first strategy includes one or more of the following:
[0013] Collaborative reasoning model;
[0014] Resource allocation information;
[0015] Task splitting point;
[0016] Number of splitting points;
[0017] Location of the split point;
[0018] Subtask unloading location;
[0019] The order in which subtasks are executed.
[0020] Optionally, in the task collaborative reasoning method, the AI capability information of the user device includes one or more of the following:
[0021] Static capability information;
[0022] Dynamic capability information.
[0023] Optionally, in the aforementioned task collaborative reasoning method, the static capability information includes one or more of the following:
[0024] Sidelink communication capability information;
[0025] Cellular communication capability information;
[0026] Static computing capability information.
[0027] Optionally, in the task collaborative reasoning method, the dynamic capability information includes one or more of the following:
[0028] Dynamic computing capability information;
[0029] Connection status information;
[0030] Energy consumption and power consumption report information;
[0031] Information on collaborative computing capabilities.
[0032] Optionally, in the task collaborative reasoning method, the Sidelink communication capability information includes one or more of the following:
[0033] The second piece of information is used to indicate whether the user equipment supports Sidelink communication.
[0034] Parameters of the Packet Data Convergence Protocol (PDCP) layer in the Sidelink direct link;
[0035] RLC layer parameters of the sidelink;
[0036] The frequency bands supported by the sidelink;
[0037] Sidelink supports frequency band combinations.
[0038] Optionally, in the task collaborative reasoning method, the cellular communication capability information includes one or more of the following:
[0039] The third piece of information is used to indicate whether the user equipment supports cellular communication;
[0040] Cellular link spectrum;
[0041] Transmission power.
[0042] Optionally, in the aforementioned task collaborative reasoning method, the static computing capability information includes one or more of the following:
[0043] The third piece of information is used to indicate whether the user equipment supports AI task computation;
[0044] Processor clock frequency;
[0045] Reasoning acceleration capability;
[0046] Supported AI frameworks.
[0047] Optionally, in the task collaborative reasoning method, the dynamic computing capability information includes computing load status and / or task status.
[0048] Optionally, in the task collaborative reasoning method, the connection status information includes one or more of signal strength, latency information, link packet loss rate, and link jitter.
[0049] One embodiment of this application also provides a task collaborative reasoning method, wherein the method is executed by a first application and includes:
[0050] The requirements for obtaining the first task, and the first information sent by the first device on the wireless access network side; wherein the first information includes one or more of the following:
[0051] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0052] AI capability information for at least one network device;
[0053] The first strategy for the first task;
[0054] Based on the requirements of the first task and / or the first information, determine a second strategy for the first task.
[0055] Optionally, in the task collaborative reasoning method, the first strategy and / or the second strategy each include one or more of the following:
[0056] Collaborative reasoning model;
[0057] Resource allocation information;
[0058] Task splitting point;
[0059] Number of splitting points;
[0060] Location of the split point;
[0061] Subtask unloading location;
[0062] The order in which subtasks are executed.
[0063] Optionally, in the task collaborative reasoning method, the requirements of the first task include one or more of the following information:
[0064] Link connection quality requirements;
[0065] Link latency requirements;
[0066] Packet loss tolerance;
[0067] Computing power threshold for collaborative devices;
[0068] A tool library that supports AI tasks;
[0069] Computational power requirements;
[0070] Minimum GPU performance;
[0071] Service time requirements;
[0072] Task priority.
[0073] Optionally, in the task collaborative reasoning method, the AI capability information of the user device includes one or more of the following:
[0074] Static capability information;
[0075] Dynamic capability information.
[0076] Optionally, in the aforementioned task collaborative reasoning method, the static capability information includes one or more of the following:
[0077] Sidelink communication capability information;
[0078] Cellular communication capability information;
[0079] Static computing capability information.
[0080] Optionally, in the task collaborative reasoning method, the dynamic capability information includes one or more of the following:
[0081] Dynamic computing capability information;
[0082] Connection status information;
[0083] Energy consumption and power consumption report information;
[0084] Information on collaborative computing capabilities.
[0085] Optionally, in the task collaborative reasoning method, the Sidelink communication capability information includes one or more of the following:
[0086] The second piece of information is used to indicate whether the user equipment supports Sidelink communication.
[0087] PDCP layer parameters of the straight-through link (Sidelink);
[0088] RLC layer parameters of the sidelink;
[0089] The frequency bands supported by the sidelink;
[0090] Sidelink supports frequency band combinations.
[0091] Optionally, in the task collaborative reasoning method, the cellular communication capability information includes one or more of the following:
[0092] The third piece of information is used to indicate whether the user equipment supports cellular communication;
[0093] Cellular link spectrum;
[0094] Transmission power.
[0095] Optionally, in the aforementioned task collaborative reasoning method, the static computing capability information includes one or more of the following:
[0096] The third piece of information is used to indicate whether the user equipment supports AI task computation;
[0097] Processor clock frequency;
[0098] Reasoning acceleration capability;
[0099] Supported AI frameworks.
[0100] Optionally, in the task collaborative reasoning method, the dynamic computing capability information includes computing load status and / or task status.
[0101] Optionally, in the task collaborative reasoning method, the connection status information includes one or more of signal strength, latency information, link packet loss rate, and link jitter.
[0102] One embodiment of this application also provides a task collaborative reasoning apparatus, wherein the apparatus is applied to a first device and includes:
[0103] A sending module is configured to send first information to a first application; wherein the first information is used by the first application to determine a second strategy for a first task, and the first information includes one or more of the following:
[0104] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0105] AI capability information for at least one network device;
[0106] The first strategy for the first task.
[0107] One embodiment of this application also provides a task collaborative reasoning device, wherein, applied to a first application, the device includes:
[0108] The acquisition module is used to acquire the requirements of the first task and to acquire first information sent by the first device on the wireless access network side; wherein the first information includes one or more of the following:
[0109] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0110] AI capability information for at least one network device;
[0111] The first strategy for the first task;
[0112] The determining module is configured to determine a second strategy for the first task based on the requirements of the first task and / or the first information.
[0113] One embodiment of this application also provides a network device, which includes: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the task collaborative reasoning method as described in any of the preceding claims.
[0114] One embodiment of this application also provides a readable storage medium, comprising: a program stored on the readable storage medium, wherein the program, when executed by a processor, implements the steps of the task collaborative reasoning method as described in any of the preceding claims.
[0115] One embodiment of this application also provides a computer program product, which includes computer instructions that, when executed by a processor, implement the steps of the task collaborative reasoning method as described in any of the preceding claims.
[0116] At least one of the above-mentioned technical solutions of this application has the following beneficial effects:
[0117] The task collaborative reasoning method described in this application provides a first application with one or more first information, including AI capability information of user devices, AI capability information of network devices, and a first strategy for the first task. This enables the first application to determine a second strategy for model segmentation and computational offloading of the first task based on the obtained first information. This allows the application to determine an effective computational offloading strategy based on communication and computing resources, thereby achieving a reasonable allocation of AI tasks, avoiding overload of individual devices, and improving the overall system's computing power and response speed. Attached Figure Description
[0118] Figure 1 This is one of the structural schematic diagrams of a collaborative reasoning mode using the task collaborative reasoning method described in the embodiments of this application;
[0119] Figure 2 This is a second schematic diagram of the collaborative reasoning mode using the task collaborative reasoning method described in the embodiments of this application;
[0120] Figure 3 This is the third schematic diagram of the collaborative reasoning mode using the task collaborative reasoning method described in the embodiments of this application;
[0121] Figure 4 This is a flowchart illustrating the task collaborative reasoning method described in Embodiment 1 of this application;
[0122] Figure 5 This is a flowchart illustrating a specific implementation of the method described in the embodiments of this application;
[0123] Figure 6 This is a flowchart illustrating the task collaborative reasoning method described in Embodiment 2 of this application;
[0124] Figure 7 This is a schematic diagram of the task collaborative reasoning device described in Embodiment 1 of this application;
[0125] Figure 8 This is a schematic diagram of the task collaborative reasoning device described in Embodiment 2 of this application. Detailed Implementation
[0126] To make the technical problems, technical solutions and advantages of this application clearer, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments.
[0127] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0128] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.
[0129] It is worth noting that the technologies described in this application are not limited to Long Term Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA), and other systems. The terms "system" and "network" in this application are often used interchangeably, and the described technologies can be used not only in the systems and radio technologies mentioned above, but also in other systems and radio technologies. The following description describes New Radio (NR) systems for illustrative purposes, and NR terminology is used in most of the following description; however, these technologies can also be applied to applications beyond NR systems, such as 6th Generation (6G) communication systems.
[0130] To achieve reasonable allocation of AI tasks under constraints of communication and computing resources, and to avoid overloading of individual devices, this application provides a task collaborative reasoning method. By providing a first application with one or more first information, including AI capability information of user devices, AI capability information of network devices, and a first strategy for the first task, the first application can determine a second strategy for model segmentation and computation offloading of the first task based on the obtained first information. This enables the first application to determine an effective computation offloading strategy based on the communication and computing resource conditions, thereby achieving reasonable allocation of AI tasks, avoiding overloading of individual devices, and improving the overall system's computing power and response speed.
[0131] To address the limitations of communication spectrum resources and terminal computing power, the task collaborative reasoning method described in this application includes at least three collaborative reasoning modes. The first collaborative reasoning mode is collaborative reasoning between the terminal and the edge server, such as... Figure 1As shown, terminal 10 initiates an AI inference task. After completing at least one layer of neural network inference before the AI task segmentation point, it transmits the intermediate results to edge server 20 via the air interface. Edge server 20, after completing at least one layer of neural network inference after the AI task segmentation point, transmits the final inference result back to the terminal via the air interface. The second collaborative inference mode is device-to-device (D2D) collaborative inference, such as... Figure 2 As shown, terminal 10 initiates an AI inference task, completes at least one layer of neural network inference before the AI task segmentation point, and then transmits the intermediate results to terminal 30 via a direct sidelink link. Terminal 30, after completing at least one layer of neural network inference after the AI task segmentation point, transmits the final inference result to terminal 10 via a sidelink link. The third collaborative inference mode is a three-way collaborative inference between the terminal, terminal, and base station, as shown... Figure 3 As shown, in this collaborative inference mode, after completing at least one layer of neural network inference before the first split point of the AI task, terminal 10 transmits the intermediate result to terminal 30 through a sidelink link. After completing at least one layer of neural network inference between the first split point and the second split point, terminal 30 transmits the intermediate result to edge server 20 through an air interface. After completing at least one layer of neural network inference after the second split point, edge server 20 transmits the final inference result to terminal 30 through an air interface. Terminal 30 then transmits the final inference result back to terminal 10 through a sidelink link. Alternatively, after completing at least one layer of neural network inference after the second split point, edge server 20 directly transmits the final inference result to terminal 10 through an air interface.
[0132] like Figure 4 This is a flowchart illustrating a task collaborative reasoning method according to one embodiment of this application. The method in this embodiment is executed by a first device and includes:
[0133] S410, send first information to the first application; wherein, the first information is used by the first application to determine a second strategy for the first task, and the first information includes one or more of the following:
[0134] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0135] AI capability information for at least one network device;
[0136] The first strategy for the first task.
[0137] The task collaborative reasoning method described in this application embodiment allows a first device to send one or more of the following first information to a first application: AI capability information of a user device, AI capability information of a network device, and a first strategy for a first task. This enables the first application to determine a second strategy for model segmentation and computational offloading of the first task based on the obtained first information. This allows the application to determine an effective computational offloading strategy based on communication and computing resources, thereby achieving a reasonable allocation of AI tasks.
[0138] In this embodiment of the application, the first device can be one of the network devices or modules of the wireless access network. Optionally, the first device can be a base station or an intelligent module in the wireless access network, such as a Near-Real-Time Radio Intelligence Controller (Near-RT RIC) in an Open Radio Access Network (O-RAN).
[0139] Optionally, the first application can be described as an application server, server, or application server, or it can be one of the application modules in the server.
[0140] Optionally, in order to support collaborative computing in AI task inference, user equipment (also known as a terminal) reports AI capability information. By using the AI capability information reported by multiple user equipment, the communication and computing resources on the user equipment side can be understood.
[0141] In some embodiments, the AI capability information may optionally include computing capability information and / or reasoning capability information, etc.
[0142] In some embodiments, the AI capability information of the user device may optionally include one or more of the following:
[0143] Static capability information;
[0144] Dynamic capability information.
[0145] Optionally, the static capability information includes one or more of the following:
[0146] Sidelink communication capability information;
[0147] Cellular communication capability information;
[0148] Static computing capability information.
[0149] Optionally, the Sidelink communication capability information includes one or more of the following:
[0150] The second piece of information is used to indicate whether the user equipment supports Sidelink communication.
[0151] Parameters of the Packet Data Convergence Protocol (PDCP) layer in the Sidelink direct link;
[0152] Wireless connectivity control RLC layer parameters for the sidelink;
[0153] The frequency bands supported by the sidelink;
[0154] Sidelink supports frequency band combinations.
[0155] The Sidelink communication capability information includes a second piece of information, which indicates whether the user equipment supports the Sidelink capability, that is, whether the user equipment has the ability to communicate directly between devices.
[0156] Sidelink communication capability information includes Sidelink's PDCP layer parameters. Optionally, these Packet Data Convergence Protocol (PDCP) layer parameters include parameters for data transmission compression, encryption, and integrity protection.
[0157] Sidelink communication capability information includes sidelink Radio Link Control (RLC) layer parameters, which optionally include parameters for capabilities such as data segmentation, retransmission, and flow control.
[0158] Sidelink communication capability information includes the frequency bands supported by the sidelink. For example, the frequency band supported by the sidelink is a single band, such as Band 47 of LTE-V2X or the n77 band of NR-V2X.
[0159] Sidelink communication capability information includes the combination of frequency bands supported by Sidelink, which indicates the spectrum range and frequency bands used by user equipment when performing D2D communication. These frequency bands may differ from those of cellular links and may also operate on unallocated spectrum.
[0160] In some embodiments, the AI capability information reported by the user equipment includes static capability information, which includes cellular communication capability information. Optionally, the cellular communication capability information includes one or more of the following:
[0161] The third piece of information is used to indicate whether the user equipment supports cellular communication;
[0162] Cellular link spectrum;
[0163] Transmission power.
[0164] The third piece of information in the cellular communication capability information is a parameter indicating whether the user equipment supports cellular communication capabilities, and is used to indicate whether the user equipment has cellular communication capabilities.
[0165] The cellular link spectrum in this cellular communication capability information is used to indicate the spectrum range and frequency band used by the cellular link, which usually involves LTE or 5G NR frequency bands.
[0166] The transmit power in this cellular communication capability information is used to indicate the terminal's maximum transmit power, usually in dBm.
[0167] In some embodiments, the AI capability information reported by the user equipment includes static capability information, which includes static computing capability information.
[0168] Optionally, the static computing capability information includes one or more of the following:
[0169] The third piece of information is used to indicate whether the user equipment supports AI task computation;
[0170] Processor clock frequency;
[0171] Reasoning acceleration capability;
[0172] Supported AI frameworks.
[0173] Among them, the third piece of information in the static computing capability information is a parameter used to indicate whether the user equipment has computing capabilities.
[0174] The processor clock frequency in static computing capability information indicates the clock frequency of the user device's processor, typically in GHz.
[0175] The inference acceleration capability in static computing capability information includes the specifications of inference acceleration AI chips (such as Neural-network Processing Unit, NPU, Graphics Processing Unit, GPU, etc.).
[0176] Supported AI frameworks are used to instruct the user device's AI model to run on the framework, determining parameters for the flexibility and performance of inference tasks, such as TensorFlow Lite, ONNX Runtime, etc.
[0177] Optionally, static computing capability information may also include the maximum number of supported parallel computing threads, which indicates the number of parallel computing threads (or computing cores) supported by the device, i.e., parameters that affect the concurrency capabilities of multitasking and inference tasks.
[0178] In some embodiments, the AI capability information reported by the user equipment may optionally include dynamic capability information to adapt to changes in capabilities over time (such as increases or decreases in device load) after the user equipment has registered. This embodiment, by introducing a dynamic update mechanism, enables the user equipment to periodically report capability changes.
[0179] Optionally, in addition to monitoring and reporting dynamic changes in information such as connection quality and / or signal strength, the user equipment may also monitor and report dynamic changes in one or more of the following: computing power, load status, and device energy consumption.
[0180] In this embodiment of the application, the dynamic capability information may optionally include one or more of the following:
[0181] Dynamic computing capability information;
[0182] Connection status information;
[0183] Energy consumption and power consumption report information;
[0184] Information on collaborative computing capabilities.
[0185] In some embodiments, the dynamic computing power information may optionally be described as a dynamic computing power report, which may include real-time computing load status and / or task status.
[0186] In some embodiments, the connection status information may optionally be described as connection quality and latency sensitivity, which may include one or more of signal strength, latency information, link packet loss rate and link jitter, that is, signal strength, latency requirements for specific inference tasks, and link stability reports (such as packet loss rate, jitter, etc.).
[0187] In some embodiments, the energy and power reporting information may optionally include remaining power and energy consumption estimates. Especially in mobile scenarios, when a user device is about to enter a low-power state, the energy and power reporting information submitted by the user device allows the application server to re-plan task allocation based on energy consumption information, avoiding excessive battery drain.
[0188] Optionally, the collaborative computing capability information refers to the ability of a user device to collaborate with other user devices when the user device is linked to other user devices via Sidelink (such as the current Sidelink bandwidth, collaborative computing load, etc.).
[0189] In one embodiment of this application, optionally, the user equipment can report its AI capability information to the base station, and the base station reports the AI capability information to the first device, which then opens the user equipment's AI capability information to the first application; in another embodiment, optionally, the first device is also the base station, and after the user equipment reports its AI capability information to the base station (first device), the base station (first device) opens the user equipment's AI capability information to the first application.
[0190] By employing the task collaborative reasoning method described in the embodiments of this application, by opening up the AI capability information of the user equipment on the wireless access network side and / or the AI capability information of the network equipment to the first application (application server), it can be ensured that the application server obtains the latest capability information of the application equipment and the network equipment.
[0191] In some embodiments, the AI capability information of the network device includes one or more of the following: computing power and load status.
[0192] Optionally, the task collaborative reasoning method described in this application embodiment may further include:
[0193] The first device calculates a first evaluation index corresponding to the AI processing capability based on the user device's AI processing capability; wherein, the user device's AI processing capability information includes the first evaluation index.
[0194] Compared to uploading raw data from user devices to indicate AI processing capabilities, such as computing power and load, a comprehensive score is generated by standardizing the weights of various parameters in the raw data of AI processing capabilities to obtain the primary evaluation index. This avoids the problem of potentially large processing overhead on servers that may result from directly using raw data for resource allocation, thus assisting application servers in making more intelligent decisions.
[0195] In some embodiments, optionally, when calculating the first evaluation index corresponding to AI processing capability, various parameters can be standardized with weights to generate a comprehensive score. The comprehensive score of various parameters includes one or more of the following: computing capability score, connection quality score, and device load index.
[0196] Taking computing power scoring as an example, the CPU and GPU parameters of each user device are weighted and summed to calculate the computing power score. For example: Computing power score = α * number of CPU cores + β * CPU clock speed + γ * number of CUDA cores + δ * GPU frequency. Where α, β, γ, and δ are the standardized weights for the number of CPU cores, CPU clock speed, number of CUDA cores, and GPU frequency, respectively.
[0197] Based on the calculation principle of the above-mentioned computing capability score, similarly, connection quality score and device load index can be calculated, which will not be illustrated here.
[0198] Alternatively, another implementation of calculating the first evaluation metric can be to directly generate the user equipment's current Unified Performance Index (UPI). That is, the first evaluation metric includes the UPI, which incorporates multiple factors, such as computing power score, connection quality score, and load index. The processed abstract data is then sent to the application server to facilitate rapid resource allocation decisions.
[0199] For example, one way to represent UPI is: UPI = α * Computing power score + β * Connectivity quality score - γ * Device load index, where α, β, and γ are the weight values corresponding to the computing power score, connectivity quality score, and device load index, respectively. The weight value corresponding to each evaluation factor (computing power score, connectivity quality score, and device load index) can be dynamically adjusted according to actual task requirements.
[0200] In another embodiment of this application, optionally, the AI capability information in the first information sent by the first device to the first application may also be raw data representing the AI processing capability of the user device, which is used by the first application to calculate the first evaluation index using at least one of the above methods.
[0201] In another embodiment, optionally, the first device may also generate a second evaluation index to assess the AI processing capability of the network device based on the network device's AI processing capability. The first information sent by the first device to the first application includes the second evaluation index related to the network device's AI capability information.
[0202] Similar to the calculation principle of the first evaluation indicator, the second evaluation indicator can also include the calculation ability scores of multiple different factors, which will not be illustrated here.
[0203] In one embodiment of this application, optionally, the first information sent by the first device to the first application includes the first strategy of the first task.
[0204] Optionally, the first strategy can also be described as an AI collaborative reasoning strategy or an AI collaborative reasoning suggestion strategy.
[0205] In this implementation, the first device can generate decision suggestions and / or application preference information for task segmentation of the first task, enabling the first application to determine a second strategy for executing the first task based on the first strategy generated by the first device, and to perform AI collaborative reasoning for the first task.
[0206] In some embodiments, the first strategy and the second strategy each include one or more of the following:
[0207] Collaborative reasoning model;
[0208] Resource allocation information;
[0209] Task splitting point;
[0210] Number of splitting points;
[0211] Location of the split point;
[0212] Subtask unloading location;
[0213] The order in which subtasks are executed.
[0214] The first device obtains the first strategy by using an optimization algorithm (such as a genetic algorithm, ant colony algorithm, and / or particle swarm algorithm) based on the AI capability information of each user device in the network, the AI capability information of at least one network device (such as the connection status information of network devices in the access network), and the requirements of the first task to be processed (including the number of layers, computational load, and / or latency requirements of the first task).
[0215] Optionally, the optimization objective of the optimization algorithm can be a single-objective optimization such as minimizing inference latency or maximizing system throughput, or a joint multi-objective optimization.
[0216] For example, the first device performs inference on the EfficientNet-B7 model (the first task), and the optimization goal of the first task is to achieve an accuracy of 95% and a latency requirement of less than 50ms.
[0217] In one embodiment, optionally, the user equipment that proposes the first task can report the requirements of the first task to the first application, and the first application can then send the requirements of the first task to the first equipment.
[0218] The requirements for the first task are the link quality requirements, computing power thresholds of collaborating devices, service time, and task priority, which are used to indicate the inference requirements for performing the first task.
[0219] Optionally, the requirements for the first task may include one or more of the following information:
[0220] Link connection quality requirements;
[0221] Link latency requirements;
[0222] Packet loss tolerance;
[0223] Computing power threshold for collaborative devices;
[0224] A tool library that supports AI tasks;
[0225] Computational power requirements;
[0226] Minimum GPU performance;
[0227] Service time requirements;
[0228] Task priority.
[0229] In some embodiments, the requirements of the first task include link connectivity quality requirements, which are used to determine the minimum performance standards of the communication links between devices to ensure the efficient execution of collaborative tasks.
[0230] In some embodiments, the requirements of the first task include a link latency requirement, which indicates the time required for data to travel from the sender to the receiver, typically in milliseconds (ms).
[0231] In some embodiments, the requirements of the first task include a packet loss tolerance, which indicates the proportion of data packets that can be lost during transmission, typically expressed as a percentage.
[0232] In some embodiments, the requirements for the first task include a computing power threshold for collaborative devices, used to screen terminal devices participating in the collaborative inference task, whose computing power must meet the minimum requirements to ensure that they can perform AI inference tasks.
[0233] In some embodiments, the requirements of the first task include computing power requirements, which are comprehensive evaluation indicators representing the overall computing power of the device, typically consisting of CPU frequency, GPU performance, memory capacity, etc. Optionally, the computing power requirements can also be expressed in the form of the computing power score described above.
[0234] In some embodiments, the requirement for the first task includes a tool library supporting AI tasks, which refers to software tools or frameworks that can help developers efficiently perform AI model inference tasks on different hardware platforms. These libraries typically provide optimized algorithms and interfaces to accelerate the model inference process, reduce latency, and improve computational efficiency, and include, for example, TensorRT, OpenVINO, etc.
[0235] In some embodiments, the requirements for the first task include minimum graphics processing unit (GPU) performance, which represents the minimum performance requirements of the GPU and is typically measured in floating-point operations per second (FLOPS) or other computational performance metrics.
[0236] In some embodiments, the requirements of the first task include a service time requirement, which indicates the maximum time interval from the start of task execution to the return of the result, typically in seconds or milliseconds.
[0237] In some embodiments, the requirements of the first task include task priority, which is used to characterize the urgency or importance of the task. For example, 0-5 can be used to characterize 5 priority levels from urgent to non-urgent.
[0238] For example, the model information (requirements for the first task) of the model used may include one or more of the following:
[0239] Model type: EfficientNet-B7 (as a target inference model);
[0240] Number of layers: 67 (including convolutional layers, activation layers, pooling layers, etc.);
[0241] Computational complexity: Approximately 66 million parameters;
[0242] Latency requirement: Inference latency less than 50ms;
[0243] Accuracy requirement: at least 95% classification accuracy.
[0244] When generating inference pattern suggestions, the first device uses an optimization algorithm (such as particle swarm optimization or genetic algorithm) to generate inference pattern suggestions based on the following factors:
[0245] Equipment capability information: UE computing power, Sidelink link bandwidth, base station computing power, etc.
[0246] Task Information: Model Type: EfficientNet-B7 (as the target inference model); Number of Layers: 67 layers (including convolutional layers, activation layers, pooling layers, etc.); Computational Cost: Approximately 66 million parameters; Latency Requirement: Inference latency less than 50ms; Accuracy Requirement: At least 95% classification accuracy.
[0247] Connection quality: This includes Sidelink link quality and cellular link quality.
[0248] Energy consumption information: The battery status of the UE device. If the battery is low, the computing load on the UE should be reduced as much as possible.
[0249] Based on this information, the optimization algorithm of the first device generates the following first strategy:
[0250] Recommended reasoning mode: UE-UE-BS collaborative reasoning mode.
[0251] Task splitting recommendation: 2 splitting points. Assign the first 10 layers of tasks of the EfficientNet-B7 model to UE1, and send the intermediate results to UE2. Execute layers 11 to 40, and send the intermediate results to the base station. The last 27 layers are processed by the base station.
[0252] It should be noted that the specific implementation of the first strategy generated by the first device described above is only an example and is not limited to this. Each implementation will not be described in detail here.
[0253] Using the above implementation method, the inference mode suggestion (first strategy) sent by the first device to the first application includes one or more of the following: recommended mode, task splitting point decision suggestion, and preliminary resource allocation configuration.
[0254] Specifically, the first strategy includes a collaborative reasoning mode, which is used to represent the reasoning mode of the selected current task scenario. This reasoning mode includes UE-BS, UE-UE, and UE-UE-BS, etc.
[0255] Optionally, the first strategy includes task splitting points, which refer to the task splitting locations in AI inference tasks where the deep neural network model is divided into different parts or layers and executed on different devices.
[0256] Optionally, the first strategy includes the number of split points, which represents the number of devices distributed across the task execution. For example, when the determined collaborative reasoning mode of the first task is UE-BS mode or UE-UE mode, the number of split points is 1, while when the determined collaborative reasoning mode of the first task is UE-UE-BS mode, the number of split points is two.
[0257] Optionally, the first strategy includes a split point location, which is used to indicate the specific layer where the execution model of the first task is split, wherein the splitting corresponds to different computational requirements in different layers, and the number of split point locations corresponds to the number of split points.
[0258] Optionally, the first strategy includes a subtask unloading location, which indicates the unloading location of the split subtasks.
[0259] Optionally, the first strategy includes a subtask execution order, which is used to indicate the execution order of the split subtasks.
[0260] Using this implementation method, the first strategy generated by the first device covers multiple aspects such as task segmentation and unloading, so as to ensure that in the D2D-assisted AI inference scenario, the allocation and execution of tasks are optimized according to dynamic conditions such as network connection, computing power, and task requirements.
[0261] Optionally, after obtaining the requirements of the first task and the first information sent by the first device on the wireless access network side (including one or more of the AI capability information of at least one user device, the AI capability information of at least one network device, and the first strategy of the first task), the first application determines the second strategy of the first task based on the requirements of the first task and / or the first information.
[0262] Optionally, the second strategy includes one or more of the following:
[0263] Collaborative reasoning model;
[0264] Resource allocation information;
[0265] Task splitting point;
[0266] Number of splitting points;
[0267] Location of the split point;
[0268] Subtask unloading location;
[0269] The order in which subtasks are executed.
[0270] The specific meaning of the above content in the second strategy is the same as that of the same content included in the first strategy, and will not be repeated here.
[0271] In this implementation, the first application determines a second strategy for the first task based on one or more of the following: AI capability information of at least one user device, AI capability information of at least one network device, the requirements of the first task, and a first strategy determined by the first device.
[0272] Optionally, the first application uses one or more of the following: AI capability information of at least one user device, AI capability information of at least one network device, the requirements of the first task, and the first strategy determined by the first device. It then uses an intelligent optimization algorithm to perform calculations, such as heuristic algorithms, genetic algorithms, or reinforcement learning algorithms, to determine a second strategy and allocates the final determined second strategy to the corresponding user terminal or base station.
[0273] Alternatively, according to the second strategy, the first application may request the network to adjust the Sidelink resource allocation.
[0274] In some embodiments, optionally, the first application utilizes AI capability information of at least one user device, AI capability information of at least one network device, and the requirements of the first task, and employs an intelligent optimization algorithm to perform multiple iterative calculations to obtain multiple candidate allocation schemes for the second strategy. For each candidate allocation scheme, the first application continues to calculate evaluation parameters for executing the first task using the corresponding candidate allocation scheme. Optionally, these evaluation parameters include capability scores, task completion time, and / or resource consumption. The evaluation parameters for executing the first task using multiple rounds of candidate allocation schemes are compared, and the optimal scheme is selected from the multiple rounds of candidate allocation schemes.
[0275] For example, suppose a user equipment (UE) has a computing power score of 70 and a link latency of 5ms, while another UE has a computing power score of 50 and a link latency of 8ms. Using an intelligent optimization algorithm, the inference tasks of layers 1-10 can be assigned to the former, and the inference tasks of layers 11-18 can be assigned to the latter, in order to maximize efficiency.
[0276] In some embodiments, optionally, when the first device provides a first strategy to the first application, after the first application performs multiple iterations of calculation using an intelligent optimization algorithm to obtain multiple candidate allocation schemes for the second strategy, the first application executes the evaluation parameters of the first task according to each candidate allocation scheme, and combines the first strategy provided by the first device to select the optimal candidate allocation scheme from the multiple candidate allocation schemes that corresponds to the first strategy as the scheme provided by the second strategy.
[0277] For example, let's take the optimization algorithm of the first device mentioned above as an example to generate the following first strategy:
[0278] Recommended reasoning mode: UE-UE-BS collaborative reasoning mode.
[0279] Task splitting recommendation: 2 splitting points. Assign the first 10 layers of tasks of the EfficientNet-B7 model to UE1, and send the intermediate results to UE2. Execute layers 11 to 40, and send the intermediate results to the base station. The last 27 layers are processed by the base station.
[0280] The requirements for the first task include:
[0281] Link latency requirement: The data transmission time from the sender to the receiver must be 20ms;
[0282] Packet loss tolerance: Allows for the loss of 0.50% of data packets during transmission;
[0283] Collaborative device computing power threshold: The terminal devices participating in collaborative inference tasks must have a computing power of 1 TFLOPS.
[0284] The first application makes a final decision and determines the second strategy based on the first strategy provided by the first device and in combination with the above-mentioned requirements of the first task.
[0285] Optionally, the decision-making process for determining the second strategy in the first application can be as follows:
[0286] Inference mode selection: Confirm the adoption of the UE-UE-BS collaborative inference mode recommended by the first device;
[0287] Task segmentation: According to the recommendations in the first strategy, the first application confirms the task segmentation point, UE1 executes the first 10 layers, UE2 executes the middle 40 layers, and the base station executes the last 27 layers;
[0288] Resource allocation: Based on the resource allocation recommendations in the first strategy and the requirements of the first task, it is confirmed that allocating 40% of cellular resources to the Sidelink link can meet the requirements of the inference task.
[0289] Based on the determined second strategy described above, the reasoning task is executed:
[0290] UE performs the following tasks:
[0291] UE1: Performs the computation of the first 10 layers of the EfficientNet-B7 model and transmits the intermediate results to UE2 via Sidelink.
[0292] UE2: Executes layers 11 to 40 of the EfficientNet-B7 model and transmits the intermediate results to the base station via the cellular link.
[0293] Base station performs the following tasks:
[0294] After receiving the intermediate results from UE2, the base station continues to execute the last 27 layers of the EfficientNet-B7 model, completes the inference calculation, and returns the final results to the application server (first application) / UE1.
[0295] By employing the method described in the embodiments of this application, by providing the first application with one or more of the following first information: AI capability information of the user device, AI capability information of the network device, and a first strategy for the first task, the first application can determine a second strategy for model segmentation and computational offloading of the first task based on the obtained first information. This allows for the determination of an effective computational offloading strategy based on communication and computing resource availability, achieving reasonable allocation of AI tasks to avoid overloading of individual devices and improve the overall system's computing power and response speed. For example, if the first task is assigned to a device with poor connection quality or insufficient computing power, the first application can determine the second strategy to adjust task allocation, inference mode, and / or computing resources to ensure that the needs of the first task are met.
[0296] In one embodiment of this application, optionally, during the execution of the first task by the user equipment or base station, the task progress and equipment load of the first task are fed back to the network (first equipment or base station) in real time.
[0297] The first device or application can allocate tasks or adjust Sidelink resources based on the task progress and device load of the first task. Additionally, if a UE's load becomes too heavy during task execution, some tasks can be reassigned to other devices to ensure the task is completed in the shortest possible time.
[0298] Taking the first device in this application as an example, which is a control device independent of the wireless access network side, combined with Figure 5 As shown, the specific implementation process of one embodiment of the task collaborative reasoning method described in this application includes the following steps:
[0299] S501, the first device collects AI capability information of at least one user device and / or AI capability information of at least one network device; wherein, the information included in the AI capability information of the user device and the AI capability information of the network device are described in detail above and will not be repeated here.
[0300] In the case where the first device is a control device on the wireless access network side that is independent of the base station, the first device may optionally obtain the AI capability information of the at least one user device and / or the AI capability information of at least one network device through the base station.
[0301] S502, the first device performs data preprocessing on the AI capability information of at least one user device and / or at least one network device obtained, optimizes the format and content of the information, and optionally calculates a first evaluation index corresponding to the AI processing capability of the user device, and / or calculates a second evaluation index corresponding to the AI processing capability of the network device; it should be noted that this step is optional and not mandatory.
[0302] S503, the first device opens the AI capability information of at least one user device and / or at least one network device to the first application to ensure that the first application obtains the latest capability information of the user device and the network device;
[0303] S504, the user equipment reports the requirements of the first task to the first application; it should be noted that this step S504 is not limited to being executed after step S503, but can also be executed synchronously with the process of the first application obtaining AI capability information of at least one user equipment and / or AI capability information of at least one network device; the specific content included in the requirements of the first task is described in detail above, and will not be repeated here.
[0304] S505, the first application sends a first request to the first device to request the first device to execute a first strategy for the first task;
[0305] S506, the first device generates a first strategy based on one or more of the following: AI capability information of at least one user device, AI capability information of at least one network device, and the requirements of the first task; wherein the first strategy generated by the first device is suggestion information for model segmentation and computational offloading of the first task, and is used as a reference for the first application when generating a second strategy. The method by which the first device generates the first strategy is described in detail above and will not be repeated here.
[0306] S507, the first device transmits the generated first policy to the first application.
[0307] It should be noted that step S505 is not a mandatory step in the embodiments of this application. When generating the first policy, the first device may not be limited to starting it based on the first request sent by the first application. After obtaining the AI capability information of at least one user device, the AI capability information of at least one network device, and the requirements of the first task, the first device actively generates the first policy based on the obtained information and transmits it to the first application.
[0308] In other embodiments, the second strategy for the first application to generate the first task for inference may not necessarily depend on the first strategy of the first device. Therefore, the above steps S505 to S507 may also be optional steps in the method described in the embodiments of this application.
[0309] S508, the first application determines a second strategy for model segmentation and computation offloading of the first task based on one or more of the user equipment's AI capability information, the network equipment's AI capability information, the first strategy of the first task, and the requirements of the first task.
[0310] S509, the first application sends task allocation information generated based on the second strategy to the user equipment and / or base station, so that the user equipment and / or base station respectively determine their cooperative roles and task allocation methods in the inference task.
[0311] S510, Optionally, if the network resources need to be adjusted for the inference of the second task according to the second strategy, the first application sends a Sidelink resource allocation optimization request to the base station to request resource allocation adjustment so as to ensure that the inference task can be executed smoothly and avoid network congestion or delay.
[0312] In some embodiments, optionally, the user equipment sends AI capability information to the base station through the RRC connection establishment process.
[0313] In one embodiment, optionally, a field is added to the UE CapabilityInformation in the terminal capability reporting signaling to report AI capability information. This added field may include CPU / GPU performance, memory size, and AI accelerator support, respectively, to report the corresponding AI capability information of the user equipment.
[0314] Optionally, after parsing the terminal capability reporting signaling, the base station encapsulates the data in a Next-Generation Application Protocol (NGAP) message via the N2 interface and forwards it to the Access and Mobility Management Function (AMF) entity. The AMF stores the computing capability data in the Unified Data Management (UDM) database and exposes it to the first device through the Network Exposure Function (NEF).
[0315] In one embodiment, optionally, if the user equipment meets the first condition, dynamic capability information reporting is triggered, and the user equipment updates its computing capability status.
[0316] Optionally, the first condition includes one or more of the following:
[0317] If the user's device has a battery level lower than a preset value, such as below 20%;
[0318] The user's device temperature exceeds the preset temperature value.
[0319] In some embodiments, under the Open Radio Access Network (O-RAN) architecture, user equipment (UE) reports computing capability parameters to the Open Distributed Unit (O-DU) via extended Radio Resource Control (RRC) signaling (such as UE Capability Information or a newly defined UE ComputeCapability Report). After receiving the UE's UE Capability Information or UE ComputeCapability Report via RRC signaling, the O-DU converts it into a computing capability report model of the E2 interface and reports the computing capability parameters to the first device via the Report Service message of the E2 interface.
[0320] In some embodiments, for application-layer extended computing power reporting, the HTTP Header / Body extension method can be used to attach computing power parameters to the business request to report the AI computing power of the user device.
[0321] In some embodiments, for the capability reporting of network devices (such as base stations) in the O-RAN architecture, optionally, the base station (such as O-DU, Open Centralized Unit (O-CU)) can directly report its computing and communication capabilities (AI capability information of the base station) to the first device (such as Near-RTRIC) by extending the E2 interface service model.
[0322] In some embodiments, for the capability reporting of network devices (such as base stations) in the O-RAN architecture, optionally, the computing capabilities in the AI capability information of the network devices can be reported by the open cloud O-Cloud to the Service Management and Orchestration (SMO) entity through the O2 interface, and the communication capability information in the AI capability information can be reported by the base station (such as O-DU, O-CU) to the SMO through the O1 interface. The SMO then reports the AI capability information of the base station to the first device.
[0323] In some embodiments, access network capabilities in the O-RAN architecture are exposed through the SMO, where the first application (or application server) sends a service discovery request to the SMO, and the SMO responds with whether access is allowed.
[0324] In some embodiments, the access network capabilities in the O-RAN architecture are exposed through the Y1 interface, where the first device (such as Near-RT RIC) exposes the collected information through the Y1 interface. This information can be exposed to external consumers, such as application servers or other network functions.
[0325] In some embodiments, decision recommendations are generated by rApp (first application) in the O-RAN architecture. rApp uses the acquired information and combines it with the application's business requirements or network optimization goals (such as reducing latency, increasing throughput, balancing load, etc.) to generate specific decision recommendations.
[0326] In some embodiments, in the O-RAN architecture, xApp (first application) generates decision recommendations, and the first device (Non-RT RIC) issues network optimization goals and / or collected information. Based on the information acquired in real time, xApp uses technologies such as machine learning and optimization algorithms to generate decision recommendations for network scheduling and resource allocation.
[0327] It should be noted that the methods for obtaining the AI capability information of user devices and network devices listed above in this application embodiment are merely illustrative examples and are not intended to limit the scope of the application.
[0328] The task collaborative reasoning method described in this application can achieve reasonable allocation of AI tasks under the constraints of communication resources and computing resources, so as to avoid the problem of overloading a single device.
[0329] One embodiment of this application also provides a task collaborative reasoning method, executed by a first application, such as... Figure 6 As shown, the method includes:
[0330] S610, obtaining the requirements of the first task, and obtaining the first information sent by the first device on the radio access network side; wherein, the first information includes one or more of the following:
[0331] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0332] AI capability information for at least one network device;
[0333] The first strategy for the first task;
[0334] S620, determine a second strategy for the first task based on the requirements of the first task and / or the first information.
[0335] The task collaborative reasoning method described in this application involves a first device providing a first application with one or more pieces of first information, including AI capability information of the user device, AI capability information of the network device, and a first strategy for the first task. This allows the first application to determine a second strategy for model segmentation and computational offloading of the first task based on the obtained first information. This enables the application to determine an effective computational offloading strategy based on communication and computing resource availability, achieving a reasonable allocation of AI tasks, avoiding overload of individual devices, and improving the overall system's computing power and response speed.
[0336] Optionally, in the task collaborative reasoning method, the first strategy and / or the second strategy each include one or more of the following:
[0337] Collaborative reasoning model;
[0338] Resource allocation information;
[0339] Task splitting point;
[0340] Number of splitting points;
[0341] Location of the split point;
[0342] Subtask unloading location;
[0343] The order in which subtasks are executed.
[0344] Optionally, in the task collaborative reasoning method, the requirements of the first task include one or more of the following information:
[0345] Link connection quality requirements;
[0346] Link latency requirements;
[0347] Packet loss tolerance;
[0348] Computing power threshold for collaborative devices;
[0349] A tool library that supports AI tasks;
[0350] Computational power requirements;
[0351] Minimum GPU performance;
[0352] Service time requirements;
[0353] Task priority.
[0354] Optionally, in the task collaborative reasoning method, the AI capability information of the user device includes one or more of the following:
[0355] Static capability information;
[0356] Dynamic capability information.
[0357] Optionally, in the aforementioned task collaborative reasoning method, the static capability information includes one or more of the following:
[0358] Sidelink communication capability information;
[0359] Cellular communication capability information;
[0360] Static computing capability information.
[0361] Optionally, in the task collaborative reasoning method, the dynamic capability information includes one or more of the following:
[0362] Dynamic computing capability information;
[0363] Connection status information;
[0364] Energy consumption and power consumption report information;
[0365] Information on collaborative computing capabilities.
[0366] Optionally, in the task collaborative reasoning method, the Sidelink communication capability information includes one or more of the following:
[0367] The second piece of information is used to indicate whether the user equipment supports Sidelink communication.
[0368] PDCP layer parameters of the straight-through link (Sidelink);
[0369] RLC layer parameters of the sidelink;
[0370] The frequency bands supported by the sidelink;
[0371] Sidelink supports frequency band combinations.
[0372] Optionally, in the task collaborative reasoning method, the cellular communication capability information includes one or more of the following:
[0373] The third piece of information is used to indicate whether the user equipment supports cellular communication;
[0374] Cellular link spectrum;
[0375] Transmission power.
[0376] Optionally, in the aforementioned task collaborative reasoning method, the static computing capability information includes one or more of the following:
[0377] The third piece of information is used to indicate whether the user equipment supports AI task computation;
[0378] Processor clock frequency;
[0379] Reasoning acceleration capability;
[0380] Supported AI frameworks.
[0381] Optionally, in the task collaborative reasoning method, the dynamic computing capability information includes computing load status and / or task status.
[0382] Optionally, in the task collaborative reasoning method, the connection status information includes one or more of signal strength, latency information, link packet loss rate, and link jitter.
[0383] The specific implementation of the method described in this application for the first application can be found in the detailed description of the specific implementation for the first device, and will not be repeated here.
[0384] One embodiment of this application also provides a task collaborative reasoning device, applied to a first device, such as... Figure 7 As shown, the device includes:
[0385] The sending module 710 is configured to send first information to a first application; wherein the first information is used by the first application to determine a second strategy for a first task, and the first information includes one or more of the following:
[0386] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0387] AI capability information for at least one network device;
[0388] The first strategy for the first task.
[0389] Optionally, the task collaborative reasoning device further includes a determining module 720, used for:
[0390] The first strategy is determined based on one or more of the AI capability information of the at least one user device, the AI capability information of the at least one network device, and the requirements of the first task.
[0391] Optionally, in the task collaborative reasoning apparatus, the first strategy includes one or more of the following:
[0392] Collaborative reasoning model;
[0393] Resource allocation information;
[0394] Task splitting point;
[0395] Number of splitting points;
[0396] Location of the split point;
[0397] Subtask unloading location;
[0398] The order in which subtasks are executed.
[0399] Optionally, in the task collaborative reasoning device, the AI capability information of the user device includes one or more of the following:
[0400] Static capability information;
[0401] Dynamic capability information.
[0402] Optionally, in the aforementioned task collaborative reasoning apparatus, the static capability information includes one or more of the following:
[0403] Sidelink communication capability information;
[0404] Cellular communication capability information;
[0405] Static computing capability information.
[0406] Optionally, in the aforementioned task collaborative reasoning device, the dynamic capability information includes one or more of the following:
[0407] Dynamic computing capability information;
[0408] Connection status information;
[0409] Energy consumption and power consumption report information;
[0410] Information on collaborative computing capabilities.
[0411] Optionally, in the aforementioned task collaborative reasoning apparatus, the Sidelink communication capability information includes one or more of the following:
[0412] The second piece of information is used to indicate whether the user equipment supports Sidelink communication.
[0413] Parameters of the Packet Data Convergence Protocol (PDCP) layer in the Sidelink direct link;
[0414] RLC layer parameters of the sidelink;
[0415] The frequency bands supported by the sidelink;
[0416] Sidelink supports frequency band combinations.
[0417] Optionally, in the aforementioned task collaborative reasoning apparatus, the cellular communication capability information includes one or more of the following:
[0418] The third piece of information is used to indicate whether the user equipment supports cellular communication;
[0419] Cellular link spectrum;
[0420] Transmission power.
[0421] Optionally, in the aforementioned task collaborative reasoning device, the static computing capability information includes one or more of the following:
[0422] The third piece of information is used to indicate whether the user equipment supports AI task computation;
[0423] Processor clock frequency;
[0424] Reasoning acceleration capability;
[0425] Supported AI frameworks.
[0426] Optionally, in the task collaborative reasoning device, the dynamic computing capability information includes computing load status and / or task status.
[0427] Optionally, in the task collaborative reasoning device, the connection status information includes one or more of the following: signal strength, latency information, link packet loss rate, and link jitter.
[0428] One embodiment of this application also provides a task collaborative reasoning device, applied to a first application, such as... Figure 8 As shown, the device includes:
[0429] The acquisition module 810 is used to acquire the requirements of the first task and to acquire first information sent by the first device on the wireless access network side; wherein the first information includes one or more of the following:
[0430] Information on the artificial intelligence (AI) capabilities of at least one user device;
[0431] AI capability information for at least one network device;
[0432] The first strategy for the first task;
[0433] The determining module 820 is configured to determine a second strategy for the first task based on the requirements of the first task and / or the first information.
[0434] Optionally, in the task collaborative reasoning apparatus, the first strategy and / or the second strategy each include one or more of the following:
[0435] Collaborative reasoning model;
[0436] Resource allocation information;
[0437] Task splitting point;
[0438] Number of splitting points;
[0439] Location of the split point;
[0440] Subtask unloading location;
[0441] The order in which subtasks are executed.
[0442] Optionally, in the task collaborative reasoning apparatus, the requirements of the first task include one or more of the following information:
[0443] Link connection quality requirements;
[0444] Link latency requirements;
[0445] Packet loss tolerance;
[0446] Computing power threshold for collaborative devices;
[0447] A tool library that supports AI tasks;
[0448] Computational power requirements;
[0449] Minimum GPU performance;
[0450] Service time requirements;
[0451] Task priority.
[0452] Optionally, in the task collaborative reasoning device, the AI capability information of the user device includes one or more of the following:
[0453] Static capability information;
[0454] Dynamic capability information.
[0455] Optionally, in the aforementioned task collaborative reasoning apparatus, the static capability information includes one or more of the following:
[0456] Sidelink communication capability information;
[0457] Cellular communication capability information;
[0458] Static computing capability information.
[0459] Optionally, in the aforementioned task collaborative reasoning device, the dynamic capability information includes one or more of the following:
[0460] Dynamic computing capability information;
[0461] Connection status information;
[0462] Energy consumption and power consumption report information;
[0463] Information on collaborative computing capabilities.
[0464] Optionally, in the aforementioned task collaborative reasoning apparatus, the Sidelink communication capability information includes one or more of the following:
[0465] The second piece of information is used to indicate whether the user equipment supports Sidelink communication.
[0466] PDCP layer parameters of the straight-through link (Sidelink);
[0467] RLC layer parameters of the sidelink;
[0468] The frequency bands supported by the sidelink;
[0469] Sidelink supports frequency band combinations.
[0470] Optionally, in the aforementioned task collaborative reasoning apparatus, the cellular communication capability information includes one or more of the following:
[0471] The third piece of information is used to indicate whether the user equipment supports cellular communication;
[0472] Cellular link spectrum;
[0473] Transmission power.
[0474] Optionally, in the aforementioned task collaborative reasoning device, the static computing capability information includes one or more of the following:
[0475] The third piece of information is used to indicate whether the user equipment supports AI task computation;
[0476] Processor clock frequency;
[0477] Reasoning acceleration capability;
[0478] Supported AI frameworks.
[0479] Optionally, in the task collaborative reasoning device, the dynamic computing capability information includes computing load status and / or task status.
[0480] Optionally, in the task collaborative reasoning device, the connection status information includes one or more of the following: signal strength, latency information, link packet loss rate, and link jitter.
[0481] It should be noted that the embodiments of this device are devices corresponding to the embodiments of the above methods. All implementations in the embodiments of the above methods are applicable to the embodiments of this device and can achieve the same technical effect.
[0482] One embodiment of this application also provides a network device, which includes a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the task collaborative reasoning method as described in any of the preceding claims.
[0483] The specific implementation of the task collaborative reasoning method, which is executed by the program running on the processor of the network device, can be found in the detailed description of the task collaborative reasoning method when applied to the first device or the first application, and will not be repeated here.
[0484] In addition, specific embodiments of this application also provide a readable storage medium storing a computer program thereon, wherein the program, when executed by a processor, implements the steps in the task collaborative reasoning method as described in any of the above.
[0485] Specifically, the readable storage medium is applied to the first device or the first application described above. When applied to the first device or the first application, the execution steps in the corresponding task collaborative reasoning method are described in detail above and will not be repeated here.
[0486] Another embodiment of the present invention also provides a computer program product, which includes computer instructions that, when executed by a processor, implement the steps in the task collaborative reasoning method as described above.
[0487] Optionally, embodiments of this application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0488] The computer program product described in this embodiment of the invention includes computer instructions that, when executed by a processor, implement the various processes of the method embodiments shown above and achieve the same technical effects. To avoid repetition, these will not be repeated here.
[0489] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0490] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can be physically included separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.
[0491] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions that cause a computer device (which may be a personal computer, server, or network device, etc.) to execute some steps of the transmission and reception methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0492] The above describes the preferred embodiments of this application. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.
Claims
1. A task coordination reasoning method, characterized in that, Performed by a first device, the method includes: Sending first information to a first application; wherein the first information is used by the first application to determine a second strategy for a first task, and the first information includes one or more of the following: Information on the artificial intelligence (AI) capabilities of at least one user device; AI capability information for at least one network device; The first strategy for the first task.
2. The task coordination reasoning method of claim 1, wherein, The method further includes: The first strategy is determined based on one or more of the AI capability information of the at least one user device, the AI capability information of the at least one network device, and the requirements of the first task.
3. The task coordination reasoning method of claim 1, wherein, The first strategy includes one or more of the following: Collaborative reasoning model; Resource allocation information; Task splitting point; Number of splitting points; Location of the split point; Subtask unloading location; The order in which subtasks are executed.
4. The task coordination reasoning method of claim 1, wherein, The AI capability information of the user equipment includes one or more of the following: Static capability information; Dynamic capability information.
5. The task coordination reasoning method of claim 4, wherein, The static capability information includes one or more of the following: Sidelink communication capability information; Cellular communication capability information; Static computing capability information.
6. The task coordination reasoning method of claim 4, wherein, The dynamic capability information includes one or more of the following: Dynamic computing capability information; Connection status information; Energy consumption and power consumption report information; Information on collaborative computing capabilities.
7. The task coordination reasoning method of claim 5, wherein, The Sidelink communication capability information includes one or more of the following: The second piece of information is used to indicate whether the user equipment supports Sidelink communication. Parameters of the Packet Data Convergence Protocol (PDCP) layer in the Sidelink direct link; RLC layer parameters of the sidelink; The frequency bands supported by the sidelink; Sidelink supports frequency band combinations.
8. The task coordination reasoning method of claim 5, wherein, The cellular communication capability information includes one or more of the following: The third piece of information is used to indicate whether the user equipment supports cellular communication; Cellular link spectrum; Transmission power.
9. The task collaborative reasoning method according to claim 5, characterized in that, The static computing capability information includes one or more of the following: The third piece of information is used to indicate whether the user equipment supports AI task computation; Processor clock frequency; Reasoning acceleration capability; Supported AI frameworks.
10. The task collaborative reasoning method according to claim 6, characterized in that, The dynamic computing capability information includes computing load status and / or task status.
11. The task collaborative reasoning method according to claim 6, characterized in that, The connection status information includes one or more of the following: signal strength, latency information, link packet loss rate, and link jitter.
12. A task collaborative reasoning method, characterized in that, Performed by a first application, the method includes: The requirements for obtaining the first task, and the first information sent by the first device on the wireless access network side; wherein the first information includes one or more of the following: Information on the artificial intelligence (AI) capabilities of at least one user device; AI capability information for at least one network device; The first strategy for the first task; Based on the requirements of the first task and / or the first information, determine a second strategy for the first task.
13. The task collaborative reasoning method according to claim 12, characterized in that, The first strategy and / or the second strategy each include one or more of the following: Collaborative reasoning model; Resource allocation information; Task splitting point; Number of splitting points; Location of the split point; Subtask unloading location; The order in which subtasks are executed.
14. The task collaborative reasoning method according to claim 12, characterized in that, The requirements of the first task include one or more of the following: Link connection quality requirements; Link latency requirements; Packet loss tolerance; Computing power threshold for collaborative devices; A tool library that supports AI tasks; Computational power requirements; Minimum GPU performance; Service time requirements; Task priority.
15. The task collaborative reasoning method according to claim 12, characterized in that, The AI capability information of the user equipment includes one or more of the following: Static capability information; Dynamic capability information.
16. The task collaborative reasoning method according to claim 15, characterized in that, The static capability information includes one or more of the following: Sidelink communication capability information; Cellular communication capability information; Static computing capability information.
17. The task collaborative reasoning method according to claim 15, characterized in that, The dynamic capability information includes one or more of the following: Dynamic computing capability information; Connection status information; Energy consumption and power consumption report information; Information on collaborative computing capabilities.
18. The task collaborative reasoning method according to claim 16, characterized in that, The Sidelink communication capability information includes one or more of the following: The second piece of information is used to indicate whether the user equipment supports Sidelink communication. PDCP layer parameters of the straight-through link (Sidelink); RLC layer parameters of the sidelink; The frequency bands supported by the sidelink; Sidelink supports frequency band combinations.
19. The task collaborative reasoning method according to claim 16, characterized in that, The cellular communication capability information includes one or more of the following: The third piece of information is used to indicate whether the user equipment supports cellular communication; Cellular link spectrum; Transmission power.
20. The task collaborative reasoning method according to claim 16, characterized in that, The static computing capability information includes one or more of the following: The third piece of information is used to indicate whether the user equipment supports AI task computation; Processor clock frequency; Reasoning acceleration capability; Supported AI frameworks.
21. The task collaborative reasoning method according to claim 17, characterized in that, The dynamic computing capability information includes computing load status and / or task status.
22. The task collaborative reasoning method according to claim 17, characterized in that, The connection status information includes one or more of the following: signal strength, latency information, link packet loss rate, and link jitter.
23. A task collaborative reasoning device, characterized in that, Applied to a first device, the device includes: A sending module is configured to send first information to a first application; wherein the first information is used by the first application to determine a second strategy for a first task, and the first information includes one or more of the following: Information on the artificial intelligence (AI) capabilities of at least one user device; AI capability information for at least one network device; The first strategy for the first task.
24. A task collaborative reasoning device, characterized in that, For a first application, the device includes: The acquisition module is used to acquire the requirements of the first task and to acquire first information sent by the first device on the wireless access network side; wherein the first information includes one or more of the following: Information on the artificial intelligence (AI) capabilities of at least one user device; AI capability information for at least one network device; The first strategy for the first task; The determining module is configured to determine a second strategy for the first task based on the requirements of the first task and / or the first information.
25. A network device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the task collaborative reasoning method as described in any one of claims 1 to 11, or implements the task collaborative reasoning method as described in any one of claims 12 to 22.
26. A readable storage medium, characterized in that, include: The readable storage medium stores a program that, when executed by a processor, implements the steps of the task collaborative reasoning method as described in any one of claims 1 to 11, or implements the steps of the task collaborative reasoning method as described in any one of claims 12 to 22.
27. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the task collaborative reasoning method as described in any one of claims 1 to 11, or implement the steps of the task collaborative reasoning method as described in any one of claims 12 to 22.
