AGI large model-based agent training method and application thereof

Abstract

The invention provides an agent training method based on an AGI large model and application thereof, and relates to the technical field of artificial intelligence, and the method is characterized in that the method comprises the following steps: 1, building a basic AGI large model, matching a corresponding data recognition receiving processing module, and through a prediction task and a comparative learning method, obtaining a data recognition receiving processing module; data content features and rules are identified, and corresponding rules are summarized; 2, combining the set task contents and simultaneously carrying out a plurality of tasks to improve the overall generalization ability; step 3, the intelligent agent accumulates corresponding processing method data and related processing result identifiers and then performs optimization in combination with the processing speed and the processing accuracy; and step 4, on the basis of the pre-trained AGI large model, performing fine adjustment on the scheme and the feature representation for different application scenes to meet the requirement of quickly adapting to new scenes and tasks. The device has the advantages that the training and using cost is reduced, popularization is convenient, meanwhile, the overall adaptability and accuracy are improved, and the overall working efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to an intelligent agent training method based on an AGI large model and its application. Background Art

[0002] With the rapid development of Internet technology and artificial intelligence technology, more and more enterprises, in order to further improve production efficiency and reduce communication costs between departments, have started to digitally transform the corresponding industries. By using artificial intelligence technology to accurately and quickly process big data, faster and more accurate decisions can be achieved. However, the common related technologies on the market currently rely on a large amount of training data, with a relatively high overall cost. At the same time, it is difficult to obtain high-quality labeled data. In addition, the generalization ability of intelligent agents is poor and they cannot quickly process new application scenarios, thus unable to meet the existing usage requirements. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent agent training method based on an AGI large model and its application, which solves the technical problems of high implementation cost and poor adaptability in the prior art, and achieves the technical effects of reducing the overall application implementation cost and having strong adaptability.

[0004] In order to achieve the above invention purpose, the technical solution adopted by the present invention is as follows:

[0005] An intelligent agent training method based on an AGI large model, comprising the following steps: Step 1, establish a basic AGI large model and adapt the corresponding data recognition, reception and processing module. The data recognition, reception and processing module can automatically obtain the corresponding basic data and settings through manual or Internet means, and identify the characteristics and laws of the data content through prediction tasks (such as masked language models) and contrast learning methods, and summarize the corresponding laws;

[0006] Step 2, combine the set task contents so that the intelligent agent can transfer to other tasks while learning and training one task, and perform several tasks on a single set of data simultaneously to improve the overall generalization ability;

[0007] Step 3, after the intelligent agent accumulates the corresponding processing method data and related processing result identifiers, it will be optimized in combination with the processing speed and processing accuracy. The optimization can be directionally formulated manually and imitate learning in combination with existing data examples to improve the accuracy and interpretability of decisions;

[0008] Step 4, on the basis of the pre-trained AGI large model, fine-tune the solution and feature representation for different application scenarios to meet the requirements of quickly adapting to new scenarios and tasks, and transfer the used knowledge and experience to new tasks through transfer technology to reduce the training time and cost.

[0009] As an improvement, the initial data obtained in step 1 needs to be manually marked for the positions and contents of key data. Through such annotations, the large model can automatically organize, plan, and learn the content it expresses. When reading other data, it can identify and judge based on the expressed content and the data at key positions, and then make corresponding adjustment plans. The data includes text, pictures, and audio and video files.

[0010] As an improvement, a data augmentation module can be added in step 1 to adjust the historical data content and then input it into the model training again.

[0011] As an improvement, in step 2, a graph neural network (GNN) can be used in the model architecture, and by refining and adjusting the relevant content of convolutional processing, it is possible to capture the relationship and structural information between data.

[0012] As an improvement, the intelligent agent can be integrated with RPA, and at the same time, it can also perform information interaction with enterprise resource planning systems (ERP) and customer relationship management systems (CRM).

[0013] An application of an intelligent agent training method based on an AGI large model, where the intelligent agent is combined with RPA (Robotic Process Automation) technology to achieve the automation of enterprise business processes; the intelligent agent is responsible for understanding task requirements and planning task processes, and the RPA robot is responsible for performing specific automated operations, such as data entry and file processing.

[0014] The intelligent agent supports multimodal interaction methods, including natural language, voice, image, etc., and can cooperate with users and other intelligent agents. In a customer service scenario, the intelligent agent can communicate with customers through speech recognition and natural language understanding, and at the same time analyze the picture information uploaded by customers through image recognition.

[0015] The intelligent agent displays relevant decision-making processes and results through visualization technology, enabling users to intuitively understand the decision-making basis of the intelligent agent.

[0016] The beneficial effects of the present invention are as follows: By using an intelligent agent training method that combines self-supervised learning and multi-tasks, the dependence on labeled data can be reduced, the generalization ability can be improved while reducing the overall training cost. By using a decision optimization strategy that combines reinforcement learning and imitation learning, the accuracy and interpretability of the intelligent agent's decision-making can be improved. By combining the process automation application method of integrating the intelligent agent with RPA, the efficient automation of enterprise business processes can be achieved. By using the intelligent agent application mode of multimodal interaction and collaborative work, the user experience and work efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1This is a flowchart of an agent training method based on the AGI large model of the present invention. Detailed implementation manners

[0018] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0019] It should be noted that the terms "first" and "second" in the present application are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of the present application, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically defined. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes unlisted steps or units, or optionally further includes other steps or units inherent to these processes, methods, products or devices.

[0020] As Figure 1 shown, an agent training method based on the AGI large model includes the following steps:

[0021] Step 1: Establish a basic AGI large model and adapt the corresponding data recognition, reception, and processing module. The data recognition, reception, and processing module can automatically obtain the corresponding basic data and settings manually or through the Internet, and identify the characteristics and patterns of the data content through prediction tasks (such as masked language models) and contrastive learning methods, and summarize the corresponding patterns. Step 2: Combine the set task content so that the agent can transfer to other tasks while learning and training one task, and perform several tasks on a single set of data simultaneously to improve the overall generalization ability. Step 3: After the agent accumulates the corresponding processing method data and relevant processing result identifiers, it will be optimized in combination with the processing speed and accuracy. The optimization can be directionally formulated manually and combined with existing data examples for imitation learning to improve the accuracy and interpretability of decision-making. Step 4: On the basis of the pre-trained AGI large model, fine-tune the solution and feature representation for different application scenarios to quickly adapt to new scenarios and tasks, and transfer the used knowledge and experience to new tasks through transfer technology to reduce training time and costs. The agent can gradually improve the recognition efficiency and accuracy of key content through a large number of data processing and recognition applications.

[0022] The initial data obtained in Step 1 needs to be marked manually for the key data positions and content. Through such annotations, the large model can automatically organize, plan, and learn the content it expresses. When reading other data, it can identify and judge through the expressed content and the key position data, and then make corresponding adjustments and plans. The data includes text, pictures, and audio and video files. In Step 1, a data augmentation module can be added to adjust the historical data content and put it back into model training. In Step 2, a graph neural network (GNN) can be used in the model architecture, and by refining and adjusting the relevant content of convolutional processing, it can provide the ability to capture the relationships and structural information between data. The agent can be integrated with RPA, and can also interact with enterprise resource planning systems (ERP) and customer relationship management systems (CRM). When multiple related tasks are processed simultaneously, tasks such as sentiment analysis and topic recognition can be performed on the data at the same time.

[0023] Application of an intelligent agent training method based on an AGI large model, where the intelligent agent is combined with RPA (Robotic Process Automation) technology to achieve the automation of enterprise business processes; the intelligent agent is responsible for understanding task requirements and planning task processes, and the RPA robot is responsible for performing specific automated operations, such as data entry and file processing; the intelligent agent supports multi-modal interaction methods, including natural language, voice, images, etc., and can cooperate with users and other intelligent agents. In a customer service scenario, the intelligent agent can communicate with customers through speech recognition and natural language understanding, and at the same time analyze the picture information uploaded by customers through image recognition; the intelligent agent displays relevant decision-making processes and results through visualization technology, enabling users to intuitively understand the decision-making basis of the intelligent agent. When applied in a customer service scenario, the intelligent agent can communicate with customers through speech recognition and natural language understanding, and at the same time analyze the picture information uploaded by customers through image recognition. When performing visualization display, the intelligent agent can extract the keywords and sentences that are focused on in the text classification task, as well as the regions and features that are focused on in the image recognition task.

[0024] When in use, maintenance personnel should always observe the usage trends of sensitive words or keywords, adjust the marked information for such keywords, and at the same time, when integrating other integration solutions, some details and identified keyword content can be adjusted accordingly, so as to make the overall recognition effect better.

[0025] The above are only the preferred embodiments of this invention patent and are not intended to limit this invention patent. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this invention patent shall be included within the protection scope of this invention patent.

Claims

1. An agent training method based on the AGI large model, characterized in that: The following steps are involved: Step 1: Establish a basic AGI model and adapt the corresponding data recognition receiving and processing module. The data recognition receiving and processing module can obtain the corresponding basic data and settings manually or automatically through the Internet, and identify the data content characteristics and rules through prediction tasks (such as masked language models) and comparative learning methods, and summarize the corresponding rules; Step 2: Combine the set task contents so that the agent can transfer to other tasks while learning and training one task, and perform several tasks simultaneously on a single set of data to improve the overall generalization ability; Step 3: After accumulating the corresponding processing method data and related processing result identification, the intelligent agent will optimize the processing speed and processing accuracy. The optimization can be manually formulated and imitated by combining existing data examples to improve the accuracy and explainability of the decision. Step 4: Based on the pre-trained AGI large model, fine-tune the scheme and feature representation for different application scenarios to quickly adapt to new scenarios and tasks, and transfer the used knowledge and experience to new tasks through migration technology to reduce training time and cost.

2. According to claim 1, the method for training an intelligent agent based on an AGI large model is characterized in that: The initial data obtained in step 1 needs to be manually marked with key data locations and content. The large model can automatically organize, plan and learn the content it expresses through such annotations. When reading other data, it can identify and judge through the expressed content and key location data, and then make corresponding adjustments and plans. The data includes text, pictures, and audio and video files.

3. The method for training an intelligent agent based on an AGI large model according to claim 1, characterized in that: In step 1, a data enhancement module can be added to adjust the historical data content and put it into model training again.

4. The method for training an intelligent agent based on an AGI large model according to claim 1, characterized in that: In step 2, a graph neural network (GNN) can be used in the model architecture, and the relationship and structural information between the data can be captured by refining and adjusting the convolution processing related content.

5. The method for training an intelligent agent based on an AGI large model according to claim 1, characterized in that: The intelligent agent can be integrated with RPA and can also interact with the enterprise resource planning system (ERP) and customer relationship management system (CRM).

6. An application of the agent training method based on the AGI large model according to claim 1, characterized in that: The intelligent agent is combined with RPA (Robotic Process Automation) technology to realize the automation of enterprise business processes; the intelligent agent is responsible for understanding task requirements and planning task processes, and the RPA robot is responsible for performing specific automated operations, such as data entry and file processing; The agent supports multimodal interaction, including natural language, voice, image, etc., and can work collaboratively with users and other agents. In customer service scenarios, the agent can communicate with customers through voice recognition and natural language understanding, and analyze picture information uploaded by customers through image recognition; The intelligent agent displays relevant decision-making processes and results through visualization technology, so that users can intuitively understand the decision basis of the intelligent agent.

Images