Multi-agent group decision-making method and related equipment for human-machine-object collaborative scenarios

Abstract

The present invention discloses a multi-agent group decision-making method and related equipment for human-machine-object collaborative scenarios, including: i Agents, use autoencoders to extract multimodal features of the data set; fuse the multimodal data within each agent to obtain the first fusion feature; in order to reflect that each agent participates in group decision-making, the weighted average method is used to weighted fuse each first fusion feature to obtain the second fusion feature with globality; distributed training is performed through multimodal data sets to obtain the local model of each agent, and the local model of each agent is iteratively aggregated to obtain the global model; the global model is used to analyze and make decisions on the second fusion feature to obtain the decision output result. It realizes the simultaneous protection of privacy data and multimodal fusion of agents to provide solutions for collaborative decision-making and improves the decision-making accuracy of multiple agents.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a multi-agent group decision-making method and related equipment for human-machine-object collaborative scenarios. Background Art

[0002] A multi-agent system is a system composed of mutually cooperating computing units - agents. Agent collaboration is to organize multiple single agents into a group and enable each agent to coordinate and cooperate effectively, thereby generating overall problem-solving capabilities. The multi-agent system is divided into two types: centralized and distributed. The centralized system consists of a core agent and multiple independent collaborative agents that are structurally dispersed. The core agent is responsible for the dynamic allocation of tasks and the dynamic scheduling of resources, and coordinates the competition and cooperation among the collaborative agents. This type of system is relatively easy to achieve system management, control and scheduling; in a distributed system, each agent is independent of each other, completely equal, and has no logical master-slave relationship. Each agent uses a communication network to negotiate and negotiate with each other according to the predetermined protocol, the system's goals, status and its own status, capabilities, resources and knowledge, and determines their respective tasks through consultation and negotiation, coordinates their respective behavior activities, and realizes the sharing of resources, knowledge, information and functions, and collaborates to complete common tasks to achieve the overall goal. The system has good encapsulation, fault tolerance, openness and scalability.

[0003] Autonomously forming response strategies based on environmental feedback information has always been a core topic in the field of intelligent agent research. In the past period of time, insufficient computing power and limitations in improvement strategies have led to a lack of fundamental progress in this field. In recent years, due to Moore's Law and the application of deep neural network methods, hardware performance and improvement algorithms have made great progress. With the application of neural networks in reinforcement learning tasks, the industry has continuously developed a series of relatively mature intelligent agent training strategies.

[0004] In the prior art, the more advanced multi-agent coordination problems mostly use DQN (Deep Q Network) or DDPG (Deep Deterministic Policy Gradient) as individual action strategies, and on this basis, design and plan the neural network between individuals. In the study of multi-agent coordination problems, the relatively mainstream implementation methods are CommNet (Communication NeuralNet), RIAL (Reinforced Inter-agent Learning) and DIAL (Differentiable Inter-agent Learning) proposed in 2016. The latest method developed based on the two is BiCNet proposed in 2017. CommNet assumes full connection within a certain range of agents by default, and uses the same network for multiple similar agents. The current state (hidden state) and communication information are used to derive the state of the next moment, and the information exchange is derived from the mean value of the hidden state. Its advantage is that it can make autonomous planning for the connection structure of agents according to the actual position changes, but its disadvantage is that the information uses the mean value which is too general and cannot handle multiple types of agents. In addition, RIAL and DIAL adopt a DQN-like solution in their individual behaviors, with one-way information exchange between agents and a one-way ring overall architecture. The difference between the two is that RIAL transmits the maximum value of the Q network result to an agent, while DIAL transmits all the results of the Q network. In the experiment, both can solve the real-life problem of multi-class collaboration, and DIAL shows good anti-signal interference ability. However, in dealing with the problem of rapid response in non-static environments, the performance of RIAL and DIAL is still insufficient.

[0005] In today's increasingly complex decision-making environment, people, machines, objects and other collaborative agents as group decision-makers often have a lot of sensitive information, and traditional centralized group decision-making methods are difficult to meet the agents' needs for protecting private content. In addition, the pursuit of high-quality decisions also requires the entire decision-making system and each agent to take into account as many data information as possible that may have differences in source, format, quality, etc. The resulting multimodal data fusion problem also brings challenges to group decision-making. Summary of the invention

[0006] The embodiments of the present invention provide a multi-agent group decision-making method, apparatus, computer equipment and storage medium for human-machine-object collaborative scenarios to improve the accuracy of multi-agent group decision-making.

[0007] In order to solve the above technical problems, the embodiment of the present application provides a multi-agent group decision-making method for human-machine-object collaborative scenarios, including:

[0008] For the i Agents, using autoencoders to extract datasets The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data. The dataset For intelligent agents In a multi-agent decision-making scenario, the dataset It is usually a multimodal data set consisting of one or more of image data, text data, audio data, video data, and sensor data;

[0009] Fusing the multimodal data within each of the agents to obtain a first fusion feature;

[0010] In order to reflect that each intelligent agent can participate in group decision-making, the weighted average method is used to perform weighted fusion on each of the first fusion features to obtain a global second fusion feature;

[0011] Performing distributed training on the multimodal data set to obtain a local model of each agent, and performing iterative aggregation training on the local model of each agent to obtain a global model;

[0012] The global model is used to analyze and make decisions on the second fusion feature to obtain a decision output result.

[0013] Optionally, for the i-th agent, the autoencoder is used to extract the data set The multimodal features include:

[0014] Assume that the encoding function of the autoencoder is , the decoding function is , the reconstruction error is , then the training objective of the autoencoder is:

[0015] ;

[0016] in, is the multimodal data of the agent, Represents any multimodal data including image data, text data, audio data, video data, and sensor data;

[0017] After training is completed, the multimodal data is obtained using the following formula The low-dimensional representation of:

[0018] 。

[0019] Optionally, distributed training is performed through the multimodal data set to obtain a local model of each agent, including:

[0020] Initialize a global decision model , and randomly initialize the global model parameters to , and then Sent to all agents involved in decision making;

[0021] In the intelligent body The General will receive As local model parameters, combining the multimodal data The low-dimensional representation of Train the local model by Constantly updated ;

[0022] Define a loss function Used to measure the performance of the model:

[0023] ;

[0024] in, represents the fitting loss, represents the regularization term;

[0025] Calculating the loss function Parameters The gradient is:

[0026] ;

[0027] Use gradient descent to update local model parameters :

[0028] ;

[0029] in, is the learning rate, which controls the step size of parameter update;

[0030] Through the process of gradient calculation and parameter update, the ideal local model parameters are obtained until the preset conditions are met. .

[0031] Optionally, the iterative aggregation training of the local model of each agent to obtain the global model includes:

[0032] Receive model parameters updated after training for all agents , or receive the increments of parameters updated by all agents Received a total of After the return content of each agent is processed by weight, a new global parameter can be obtained. , respectively expressed as:

[0033] or

[0034] ;

[0035] in, It is an intelligent agent The weight of

[0036] Using the updated parameters Perform aggregated iterative update training, and obtain the global model when the iteration reaches the preset conditions.

[0037] Optionally, the preset condition is that the average loss is less than a threshold: , or, the number of iterations ,in, is the global loss function threshold, T is the maximum number of model iterations.

[0038] Optionally, fusing the multimodal data within each of the agents to obtain a first fusion feature includes:

[0039] The eigenvectors for each mode Perform linear transformation to obtain query ,key Sum Three vectors:

[0040] ;

[0041] in, is a learnable weight matrix;

[0042] Compute the attention scores between each modality and the other modalities:

[0043] ;

[0044] in, Representing an Agent The query vector and agent The attention score between the key vectors of ;

[0045] Perform softmax normalization on the attention score of each row to get the attention weight

[0046] ;

[0047] The attention weights and values ​​are weighted summed to get the output vector:

[0048] ;

[0049] Set the bilinear pooling function to , then the feature vector after multimodal fusion within a single agent is:

[0050] ;

[0051] in, is the multimodal data within a single agent, They are respectively latent vectors of multimodal data composed of one or more of image data, text data, audio data, video data, and sensor data.

[0052] Optionally, the step of performing weighted fusion on each of the first fusion features by using a weighted average method to obtain a global second fusion feature comprises:

[0053] The second fusion feature is expressed by the following formula:

[0054] ;

[0055] in, It is an intelligent agent The weight is determined by the importance of the agent. is the total number of agents.

[0056] In order to solve the above technical problems, the embodiment of the present application further provides a multi-agent group decision-making device for human-machine-object collaborative scenarios, including:

[0057] Feature extraction module is used to i Agents, using autoencoders to extract datasets The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data. The dataset For intelligent agents In a multi-agent decision-making scenario, the dataset It is usually a multimodal data set consisting of one or more of image data, text data, audio data, video data, sensor data, etc.

[0058] A first fusion module is used to fuse the multimodal data within each agent to obtain a first fusion feature;

[0059] The second fusion module is used to reflect that each intelligent agent can participate in group decision-making, so the weighted average method is used to perform weighted fusion on each first fusion feature to obtain a global second fusion feature;

[0060] Iterative training module, which is used to perform distributed training through multimodal data sets to obtain local models of each agent, and iteratively aggregate the local models of each agent to obtain a global model;

[0061] The decision analysis module is used to use the global model to analyze and make decisions on the second fusion feature to obtain a decision output result.

[0062] In order to solve the above-mentioned technical problems, an embodiment of the present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned multi-agent group decision-making method for human-machine-object collaborative scenarios are implemented.

[0063] In order to solve the above technical problems, an embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned multi-agent group decision-making method for human-machine-object collaborative scenarios.

[0064] The multi-agent group decision-making method, device, computer equipment and storage medium for human-machine-object collaborative scenarios provided by the embodiments of the present invention are aimed at the first i Agents, using autoencoders to extract datasets The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data. The dataset For intelligent agents In a multi-agent decision-making scenario, the dataset It is usually a multimodal data set composed of one or more of image data, text data, audio data, video data, sensor data, etc.; the multimodal data within each agent is fused to obtain the first fusion feature; each first fusion feature is weightedly fused using the weighted average method to obtain the second fusion feature with globality; distributed training is performed on the multimodal data set to obtain the local model of each agent, and the local model of each agent is iteratively aggregated to obtain the global model; the global model is used to analyze and make decisions on the second fusion feature to obtain the decision output result. It realizes the simultaneous protection of privacy data and multimodal fusion of agents to provide solutions for collaborative decision-making and improves the decision accuracy of multiple agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0066] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0067] Figure 2 It is a flowchart of an embodiment of the multi-agent group decision-making method for human-machine-object collaborative scenarios of the present application;

[0068] Figure 3 It is a structural schematic diagram of an embodiment of a multi-agent group decision-making device for human-machine-object collaborative scenarios according to the present application;

[0069] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0071] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] See also Figure 1 ,like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0074] Users can use terminal devices 101 , 102 , 103 to interact with server 105 via network 104 to receive or send messages, etc.

[0075] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV), laptop computers, desktop computers, etc.

[0076] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0077] It should be noted that the multi-agent group decision-making method for human-machine-object collaborative scenarios provided in the embodiments of the present application is executed by a server, and accordingly, the multi-agent group decision-making device for human-machine-object collaborative scenarios is arranged in the server.

[0078] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. The terminal devices 101, 102, 103 in the embodiment of the present application may specifically correspond to application systems in actual production.

[0079] See also Figure 2 , Figure 2 A multi-agent group decision-making method for human-machine-object collaboration provided by an embodiment of the present invention is shown. Figure 1 The server in is used as an example to illustrate the details as follows:

[0080] S201: For the iAgents, using autoencoders to extract datasets The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data. The dataset For intelligent agents In a multi-agent decision-making scenario, the dataset It is usually a multimodal data set consisting of one or more of image data, text data, audio data, video data, sensor data, etc.

[0081] Specifically, the agent The data set is In a decision-making scenario composed mainly of human, machine, and object agents, the data set It is usually a multimodal data set consisting of one or more of image data, text data, audio data, video data, sensor data, etc. , use autoencoder to extract its multimodal features. The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data.

[0082] In this embodiment, by minimizing the reconstruction error, the autoencoder can learn the implicit features of multimodal data. Set the objective function It is used to measure the degree to which the autoencoder learns to represent the effective features of the input data. For continuous valued data (such as images), the commonly used objective function is the mean square error (MSE); for categorical data (such as word embeddings of text), the cross entropy loss is a more appropriate choice; for multimodal data containing multiple types, MSE and cross entropy can be combined to calculate the corresponding losses for different types of data, and then these losses are weighted and summed. In the objective function, Represents the model parameters, which include encoder parameters and decoder parameters. The former is responsible for mapping the input data to a lower-dimensional hidden layer or representation layer, while the latter is responsible for reconstructing the original data from the hidden layer representation. The agent uses optimization algorithms such as gradient descent (GD), stochastic gradient descent (SGD), etc. To minimize the objective function, the model parameters Make optimizations.

[0083] In a specific embodiment, for i Agents, using autoencoders to extract datasets The multimodal features include:

[0084] Assume that the encoding function of the autoencoder is , the decoding function is , the reconstruction error is , then the training objective of the autoencoder is:

[0085] ;

[0086] in, is the multimodal data of the agent, Respectively represent any one or more multimodal data including image data, text data, audio data, video data, and sensor data;

[0087] After training is completed, the multimodal data is obtained using the following formula The low-dimensional representation of:

[0088] 。

[0089] S202: Fusing the multimodal data within each agent to obtain a first fusion feature.

[0090] In a specific optional implementation, the multimodal data within each agent is fused to obtain the first fusion feature including:

[0091] The eigenvectors for each mode Perform linear transformation to obtain query ,key Sum Three vectors:

[0092] ;

[0093] in, is a learnable weight matrix;

[0094] Compute the attention scores between each modality and the other modalities:

[0095] ;

[0096] in, Representing an Agent The query vector and agent The attention score between the key vectors of ;

[0097] Perform softmax normalization on the attention score of each row to get the attention weight

[0098] ;

[0099] The attention weights and values ​​are weighted summed to get the output vector:

[0100] ;

[0101] Set the bilinear pooling function to , then the feature vector after multimodal fusion within a single agent is:

[0102] ;

[0103] in, is the multimodal data within a single agent, They represent latent vectors of multimodal data such as image data, text data, audio data, video data, sensor data, etc.

[0104] Furthermore, the bilinear feature matrix is ​​summed and pooled to obtain the bilinear feature vector

[0105] ;

[0106] Finally, the bilinear feature vector is subjected to signed square root and L2 normalization to obtain the final bilinear pooling output vector:

[0107] .

[0108] S203: The first fusion feature of each agent is weightedly fused by using a weighted average method to obtain a global second fusion feature.

[0109] The second fusion feature is expressed by the following formula:

[0110]

[0111] in, It is an intelligent agent The weight is determined by the importance of the agent. is the total number of agents.

[0112] S204: Distributed training is performed on a multimodal data set to obtain a local model of each intelligent agent, and then the local model of each intelligent agent is iteratively aggregated and trained to obtain a global model.

[0113] In a specific optional implementation, distributed training is performed through a multimodal data set to obtain a local model of each agent, including:

[0114] Initialize a global decision model , and randomly initialize the global model parameters to , and then Sent to all agents involved in decision making;

[0115] In the intelligent body The General will receive As the local model parameters, the local model is trained by combining the first fusion feature. Constantly updated ;

[0116] Define a loss function Used to measure the performance of the model:

[0117] ;

[0118] in, represents the fitting loss, represents the regularization term;

[0119] Calculating the loss function Parameters The gradient is:

[0120] ;

[0121] Use gradient descent to update local model parameters :

[0122] ;

[0123] in, is the learning rate, which controls the step size of parameter update;

[0124] Through the process of gradient calculation and parameter update, until the preset conditions are met, the local model parameters are obtained. .

[0125] The above training process can be expressed as:

[0126] in, It is an intelligent agent The gradient descent optimization algorithm used to train the local model parameters, It is an intelligent agent Dataset The multimodal feature representation obtained after processing by the autoencoder, is the number of rounds of local training, is the batch size for each training round, is the learning rate.

[0127] In a specific optional implementation, the local model of each agent is iteratively aggregated and trained to obtain a global model including:

[0128] Receive model parameters updated after training for all agents , or receive the increments of parameters updated by all agents Received a total of After the return content of each agent is processed by weight, a new global parameter can be obtained. , respectively expressed as:

[0129] or

[0130] ;

[0131] in, It is an intelligent agent The weight of

[0132] Using the updated parameters Perform aggregated iterative update training, and obtain the global model when the iteration reaches the preset conditions.

[0133] It should be understood that the above steps are repeated to ensure that each agent involved in decision-making repeats local training and continuously participates in assisting the central server to aggregate and update the global model.

[0134] It should be noted that It is an intelligent agent The weights of each agent can be determined based on the amount of data, data quality or other relevant factors. to ensure that parameter updates are normalized.

[0135] In a specific optional implementation manner, the preset condition is that the average loss is less than a threshold value: , or, the number of iterations ,in, is the global loss function threshold, and T is the maximum number of model iterations.

[0136] The global model is considered to have achieved the expected performance and the iteration stops. Represents the loss function of all agents In the parameters The average value of is the preset loss threshold. When the average loss is less than this threshold, the training process can be stopped. is the current iteration number. Thus, we can get a fully trained global decision model , whose parameters are .

[0137] S205: Using the global model to analyze and make decisions on the second fusion feature to obtain a decision output result.

[0138] Specifically, the central server uses the trained global model (Its parameters are ) for the fused global data To process:

[0139] ;

[0140] in It is the decision output based on the global model.

[0141] In this embodiment, for i Agents, using autoencoders to extract datasets The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data. The dataset For intelligent agents In a multi-agent decision-making scenario, the dataset It is a multimodal data set composed of one or more of image data, text data, audio data, video data, sensor data, etc.; the multimodal data within each agent is fused to obtain the first fusion feature; in order to reflect that each agent can participate in group decision-making, the weighted average method is used to weightedly fuse each first fusion feature to obtain the second fusion feature; distributed training is performed on the multimodal data set to obtain the local model of each agent, and the local model of each agent is iteratively aggregated to obtain the global model; the global model is used to analyze and make decisions on the second fusion feature to obtain the decision output result. It realizes the simultaneous protection of privacy data and multimodal fusion of agents to provide solutions for collaborative decision-making and improves the decision-making accuracy of multiple agents.

[0142] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0143] Figure 3 The principle block diagram of the multi-agent group decision-making device for human-machine-object collaborative scenarios is shown, which corresponds one-to-one with the multi-agent group decision-making method for human-machine-object collaborative scenarios in the above embodiment. Figure 3 As shown, the multi-agent group decision-making device for human-machine-object collaborative scenarios includes a feature extraction module 31, a first fusion module 32, a second fusion module 33, an iterative training module 34 and a decision analysis module 35. The functional modules are described in detail as follows:

[0144] Feature extraction module 31, for the i-th agent, its corresponding data set It is a multimodal data set consisting of one or more of image data, text data, audio data, video data, sensor data, etc. Extracting data sets using autoencoders The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder can reconstruct the representation of the low-dimensional space into the original data.

[0145] A first fusion module 32, used to fuse the multimodal data within each agent to obtain a first fusion feature;

[0146] The second fusion module 33 is used to reflect that each intelligent agent can participate in group decision-making, and the first fusion feature of each intelligent agent is weighted and fused by weighted average method to obtain the second fusion feature with globality;

[0147] An iterative training module 34 is used to perform distributed training through a multimodal data set to obtain a local model of each agent, and perform iterative aggregation training on the local model of each agent to obtain a global model;

[0148] The decision analysis module 35 is used to use a global model to analyze and make decisions on the second fusion feature to obtain a decision output result.

[0149] For the specific definition of the multi-agent group decision-making device for human-machine-object collaborative scenarios, please refer to the definition of the multi-agent group decision-making method for human-machine-object collaborative scenarios above, which will not be repeated here. Each module in the above-mentioned multi-agent group decision-making device for human-machine-object collaborative scenarios can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0150] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0151] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components connected to the memory 41, the processor 42, and the network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0152] The computer device may be a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with the user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0153] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or D interface display memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as the program code of the multi-agent group decision-making method for the human-machine-object collaborative scenario, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0154] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run the program code stored in the memory 41 or process data, such as running the program code of the multi-agent group decision-making method for human-machine-object collaborative scenarios.

[0155] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0156] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores an interface display program, and the interface display program can be executed by at least one processor so that the at least one processor performs the steps of the multi-agent group decision-making method for human-machine-object collaborative scenarios as described above.

[0157] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0158] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A multi-agent group decision-making method for human-machine-object collaborative scenarios, characterized in that: include: For the i-th agent, use the autoencoder to extract the dataset D i The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data. The dataset D i Let be the data set of agent i. In the decision-making scenario composed of multiple agents, the data set D i A multimodal data set consisting of one or more of image data, text data, audio data, video data, and sensor data; Fusing the multimodal data within each of the agents to obtain a first fusion feature; In order to reflect that each intelligent agent participates in group decision-making, a weighted average method is used to perform weighted fusion on each of the first fusion features to obtain a second fusion feature; Performing distributed training on the multimodal data set to obtain a local model of each agent, and performing iterative aggregation training on the local model of each agent to obtain a global model; Using the global model to analyze and make decisions on the second fusion feature to obtain a decision output result; Among them, for the i-th agent, the autoencoder is used to extract its corresponding data set D i The multimodal features include: Assume that the encoding function of the autoencoder is E(·), the decoding function is D(·), and the reconstruction error is L r (·), then the training objective of the autoencoder is: Among them, D i =(X i ,Y i ,…) is the multimodal data of the agent, where X i ,Y i ,…represent a kind of multimodal data among image data, text data, audio data, video data and sensor data respectively; After training is completed, the multimodal data D is obtained using the following formula i The low-dimensional representation of: Z i =E(D i ); The step of fusing the multimodal data within each of the agents to obtain the first fusion feature includes: For each mode's eigenvector Z i Perform linear transformation to obtain query Q i , key K i Sum value V i Three vectors: Q i =W q Z i K i =In k WITH i V i =W v Z i Among them, W q ,W k ,W v is a learnable weight matrix; Compute the attention scores between each modality and the other modalities: Among them, S ij represents the attention score between the query vector of agent i and the key vector of agent j; Perform softmax normalization on the attention score of each row to get the attention weight The attention weights and values ​​are weighted summed to get the output vector: Set the bilinear pooling function to B p (·), then the feature vector after multimodal fusion within a single agent is: Z' i =B p (Z ′ i ) Among them, Z′ i =(X′ i ,Y′ i ,...) is the first fusion feature obtained after the multimodal data fusion within a single agent, X′ i , Y′ i ...represents the latent vector of multimodal data corresponding to image data, text data, audio data, video data, and sensor data; The step of using the weighted average method to perform weighted fusion on each of the first fusion features to obtain the second fusion feature includes: The second fusion feature is expressed by the following formula: Among them, w i is the weight of agent i, which is determined by the importance of the agent, and N is the total number of agents.

2. The multi-agent group decision-making method for human-machine-object collaborative scenarios as claimed in claim 1, characterized in that: The distributed training is performed through the multimodal data set to obtain the local model of each agent, including: Initialize a global decision model M, and randomly initialize the global model parameters to θ0, and then send θ0 to all agents involved in the decision-making; In agent i, the received θ0 is used as the local model parameter, combined with the multimodal data D i The low-dimensional representation Z i Train the local model and continuously get updated θ through θ0 i ; Define a loss function L(θ i ) is used to measure the performance of the model: l(θi)=F(θ i ,WITH i )+R(θ i ); Among them, F(θ i ,Z i ) represents the fitting loss, R(θ i ) represents the regularization term; Calculate the loss function L(θ i ) for the parameter θ i The gradient is: Use gradient descent to update the local model parameters θ ′ i : Among them, η is the learning rate, which controls the step size of parameter update; Through the process of gradient calculation and parameter update, until the preset conditions are met, the ideal local model parameter θ is obtained ′ i .

3. The multi-agent group decision-making method for human-machine-object collaborative scenarios as claimed in claim 1, characterized in that: The iterative aggregation training of the local model of each agent to obtain the global model includes: Receive the ideal local model parameters θ obtained by all agents after training and updating ′ i , or receive the increment m of the parameters updated by all agents i =θ′ i -θ0; After receiving the return content from a total of N agents, a new global parameter θ1 can be obtained through weighted processing, which can be expressed as: or Among them, w i is the weight of agent i; The updated parameter θ1 is used to perform aggregate iterative update training, and when the iteration reaches the preset conditions, the global model is obtained.

4. The multi-agent group decision-making method for human-machine-object collaborative scenarios as claimed in claim 3, characterized in that: The preset condition is that the average loss is less than the threshold: Alternatively, the number of iterations t = T, where ∈ is the global loss function threshold and T is the maximum number of model iterations.

5. A multi-agent group decision-making device for human-machine-object collaborative scenarios, characterized in that: The multi-agent group decision-making method for human-machine-object collaborative scenarios according to any one of claims 1 to 4 is adopted, and the device comprises: The feature extraction module is used to extract the dataset D using the autoencoder for the i-th agent. i The autoencoder consists of an encoder and a decoder. The encoder maps the multimodal data to a low-dimensional space, and the decoder reconstructs the representation of the low-dimensional space into the original data. The dataset D i Let the data set corresponding to agent i be, in the decision-making scenario composed of multiple agents, the data set D i A multimodal data set consisting of one or more of image data, text data, audio data, video data, and sensor data; A first fusion module is used to fuse the multimodal data within each agent to obtain a first fusion feature; The second fusion module is used to perform weighted fusion on each first fusion feature by using a weighted average method to obtain a second fusion feature; Iterative training module, which is used to perform distributed training through multimodal data sets to obtain local models of each agent, and iteratively aggregate the local models of each agent to obtain a global model; The decision analysis module is used to use the global model to analyze and make decisions on the second fusion feature to obtain a decision output result.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the multi-agent group decision-making method for human-machine-object collaborative scenarios as described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the multi-agent group decision-making method for human-machine-object collaborative scenarios as described in any one of claims 1 to 4 is implemented.

Images