Dynamic demonstration method and system of high-order course algorithm based on generative AI

Abstract

The application relates to the technical field of network teaching, in particular to a high-order course algorithm dynamic demonstration method and system based on generative AI, which encodes algorithm parameter adjustment data in target algorithm teaching interaction into structured data; inputs the structured data and algorithm input data into a pre-trained generative AI model, dynamically generates a visual description of a target algorithm processing flow by using the generative AI model, and converts the visual description into visual instructions to realize real-time dynamic visualization of the algorithm processing flow based on the visual instructions. The application generates a visual result of the algorithm processing flow in real time, enables students to adjust parameters in real time and observe the change of the internal processing state of the algorithm, helps students understand the role of algorithm core parameters and the tuning idea, establishes systematic and visual cognition of the working principle of the algorithm, deepens the understanding of students on complex principles, and improves the flexibility and effectiveness of classroom guidance.

Description

Technical Field

[0001] This invention relates to the field of online teaching technology, and in particular to a method and system for dynamically demonstrating advanced course algorithms based on generative AI. Background Technology

[0002] Generative Artificial Intelligence (GAI) technology is penetrating the education field with unprecedented depth and breadth. Its application has moved from early theoretical discussions and sporadic tool assistance to a new stage of large-scale practice and model innovation. Currently, in the field of educational technology, especially in interdisciplinary teaching related to digital and intelligent courses and artificial intelligence education, existing technologies mainly revolve around two core directions: the automated generation of intelligent teaching resources and the interactive and dynamic presentation of the teaching process. However, these technologies still have significant structural limitations in practical applications, especially for advanced courses such as "Intelligent Image Processing and Recognition" that involve teaching complex dynamic algorithms. Existing technologies cannot achieve real-time dynamic visualization of algorithm processing flows; algorithm demonstrations mostly remain at the level of static code or flowcharts, lacking an intuitive presentation of the dynamic data flow of the algorithm's internal collaborative work.

[0003] Existing educational technology applications exhibit a situation of both isolated breakthroughs and fragmented application scenarios. On the one hand, AI-assisted lesson preparation, intelligent question answering, and code generation tools have achieved considerable popularity; on the other hand, these technologies are often applied independently to a specific stage of the teaching process, failing to connect the entire cycle of "pre-class - in-class - post-class" to form an organic closed loop. When students encounter complex algorithms such as deep learning-based object detection (R-CNN, YOLO series), their understanding process still faces a significant gap between abstract theory and concrete practice. In the pre-class preparation stage, the use of generative AI to assist or even automate the generation of lesson plans and courseware has become an important technological trend, which can be summarized as follows: Teachers provide original teaching resources such as course outlines, textbook texts, and multimedia materials. The system uses Large Language Model (LLM) and multimodal semantic parsing technology to understand the content. Based on the input content, the system extracts knowledge units and models the ontology to construct a structured knowledge graph representing course knowledge points and their relationships. This process aims to transform unstructured teaching materials into a knowledge system that machines can understand and process. Based on the knowledge graph and preset teaching design templates, the system calls long text generation technology to automatically write lesson plans and courseware scripts (including verbatim transcripts). At the same time, to enhance the intuitiveness of the courseware, the system can call external computing engines (such as WolframAlpha) or multimodal generative models to generate visual charts, diagrams, and even accompanying images for core concepts and principles. This can be achieved through an integrated intelligent teaching platform or by combining existing AI toolchains. A common practical approach for teachers is to use large-scale models such as DeepSeek, Kimi, and Wenxin Yiyan to generate content outlines and texts, and then use AI-enabled office software such as WPS AI and MindShow to convert them into PPT courseware with one click. In classroom teaching, especially in algorithm and programming courses, enhancing the interactivity and visualization of courseware is another important trend. The main applications of generative AI in this field are: GAI can dynamically generate runnable code snippets based on natural language instructions (such as "write a convolutional neural network code") and explain its logic step by step, helping students understand complex structures; AI can analyze code submitted by students, locate syntax or logic errors (such as array out-of-bounds errors) in real time, and provide modification suggestions; teachers can use AI to generate interactive web page materials (such as dynamic data curves), knowledge graphs, or 3D models to demonstrate algorithm principles, lowering the barrier to creating visualization resources; based on the analysis of student learning data, GAI can recommend differentiated exercises, cases, and even AI learning partners playing different roles in multi-agent virtual classrooms, achieving a certain degree of interaction.

[0004] Currently, in digital intelligence courses such as "Image Intelligent Processing and Recognition," although AI can generate initial drafts of courseware, problems such as unclear key points and difficulties, and low matching between text and images are common. Furthermore, the generated case code and static illustrations are difficult to dynamically adjust based on real-time feedback from different classes. Teachers still need to invest a significant amount of time in precise secondary processing and verification, resulting in limited efficiency improvements. For core algorithms such as R-CNN and YOLOv3, existing materials do not document mature dedicated teaching demonstration systems. During teaching, algorithms are broken down into isolated code modules or static flowcharts. The dynamic data flow, feature map changes, and continuous processes of detection box generation and adjustment within these algorithms cannot be demonstrated and interactively controlled in real time and intuitively, making it difficult for students to develop a systematic and visual understanding. Pre-class AI lesson preparation, in-class interactive tools, and post-class intelligent tutoring are often supported by different technologies or platforms, operating independently and creating "data silos" and "experience gaps." The lack of a unified framework or system to organically link pre-class AI-generated courseware (pre-set interaction points, test cases) with in-class dynamic algorithm demonstration engines and real-time evaluation feedback makes it impossible to form a complete teaching loop of "pre-set-generation-interaction-evaluation-optimization", which limits the overall improvement of teaching effectiveness and the true realization of personalized teaching. Summary of the Invention

[0005] To address the issues of poor adaptability between AI-generated content and dynamic teaching scenarios, lack of dynamic interpretation, and deep integration throughout the entire teaching cycle in existing digital and intelligent professional courses, this invention provides a method and system for dynamic demonstration of advanced course algorithms based on generative AI. It deeply integrates generative artificial intelligence (GAI) into the teaching design, content generation, process interaction, and effect evaluation aspects that education cares about, and generates visualized results of the algorithm processing flow in real time.

[0006] According to the design scheme provided by this invention, on the one hand, a method for dynamically demonstrating advanced course algorithms based on generative AI is provided, comprising:

[0007] Obtain the target algorithm, algorithm parameters, and algorithm input data for advanced courses, and generate teaching courseware for the target algorithm;

[0008] The system receives algorithm parameter adjustment data from interactive teaching of the target algorithm and encodes the algorithm parameter adjustment data into structured data. The structured data and algorithm input data are input into a pre-trained generative AI model. The generative AI model is used to dynamically generate a visual description of the target algorithm processing flow and convert the visual description into visual instructions to present the algorithm processing flow in real time based on the visual instructions.

[0009] As a method for dynamically demonstrating advanced course algorithms based on generative AI, this invention further generates teaching materials for the target algorithm, including:

[0010] Based on the needs of algorithm teaching, algorithm teaching courseware is generated using a generative AI model. Several interactive points are set in the algorithm teaching courseware to facilitate teaching interaction. These interactive points are used to visualize the algorithm-related processing flow by adjusting algorithm parameters during algorithm teaching interaction.

[0011] As a method for dynamically demonstrating advanced course algorithms based on generative AI, this invention further includes generating algorithm teaching materials using a generative AI model according to algorithm teaching needs, comprising:

[0012] Obtain the algorithm teaching requirements input by the user, which include the teaching objects, teaching objectives, and reference resources;

[0013] By performing multimodal semantic analysis on reference resources, key knowledge units for algorithm teaching are extracted, and a structured knowledge graph for algorithm teaching is constructed.

[0014] The system uses a structured knowledge graph of teaching objects, teaching objectives, and algorithm teaching. It leverages prompting engineering to guide the generation of algorithm teaching courseware using a large language model. Based on the algorithm teaching objectives, it generates algorithm teaching instructions and calls a code generator to generate algorithm code and code examples. Furthermore, it employs a multimodal resource model built on generative artificial intelligence to acquire multimodal teaching resources. These multimodal teaching resources include text, images, videos, and audio corresponding to the algorithm teaching courseware. The multimodal resource model includes a text generation module, an image generation model, and an audio / video generation model.

[0015] As a dynamic demonstration method for advanced course algorithms based on generative AI, this invention further utilizes prompting engineering to guide teaching materials for large language model generation algorithms, including:

[0016] Based on the teaching objects and teaching objectives, set up algorithm-structured courseware task descriptions and generate prompts and instructions;

[0017] Interact with the large language model and algorithm teaching structured knowledge graph according to the prompts to obtain structured courseware content;

[0018] Algorithm-based teaching materials are generated from structured courseware content.

[0019] As a dynamic demonstration method for advanced course algorithms based on generative AI, this invention further includes interacting with a large language model and a structured knowledge graph for algorithm teaching according to prompts, comprising:

[0020] Knowledge representation learning is performed on entities and relationships in the structured knowledge graph for algorithm teaching, and the knowledge node vectors are stored in the knowledge base.

[0021] The prompt instructions are converted into query vectors, and the similarity between the query vectors and knowledge node vectors is calculated in the knowledge base to obtain the entities most relevant to the query vectors.

[0022] Using the obtained entities as nodes, perform a depth-first traversal in the graph to find the relationship paths that meet the prompt instructions, and obtain the knowledge subgraph and related reference resource fragments.

[0023] The knowledge subgraph and related reference resource fragments are spliced ​​together and a prompt instruction template is injected to enable the large language model to generate structured courseware content based on the enhanced context.

[0024] As a method for dynamically demonstrating advanced course algorithms based on generative AI, this invention further includes, through real-time dynamic visualization of the algorithm processing flow based on visualization instructions:

[0025] Based on the algorithm teaching materials, the algorithm demonstration screen is divided into a code area, a parameter control area, and a visualization area. In the algorithm interaction, the algorithm code is displayed in the code area by an AI agent, the algorithm parameters are displayed in the parameter control area by an AI agent and the interaction is achieved by adjusting the algorithm parameters, and the algorithm processing process is visualized in the visualization area by an AI agent.

[0026] As a dynamic demonstration method for advanced course algorithms based on generative AI, this invention further includes:

[0027] Based on the interactive data, a learning profile of the algorithm is created and suggestions for optimizing algorithm teaching are obtained. These suggestions are then fed back to the generative AI model, which is used to optimize the algorithm teaching materials using a combination of cloud inference and edge computing.

[0028] Furthermore, this invention also provides a dynamic demonstration system for advanced course algorithms based on generative AI, comprising: a courseware generation module and a dynamic interaction module, wherein...

[0029] The courseware generation module is used to obtain the target algorithm, algorithm parameters, and algorithm input data for advanced courses, and generate teaching courseware for the target algorithm.

[0030] The dynamic interaction module is used to receive algorithm parameter adjustment data in the teaching interaction of the target algorithm and encode the algorithm parameter adjustment data into structured data. The structured data and algorithm input data are input into a pre-trained generative AI model. The generative AI model is used to dynamically generate a visual description of the target algorithm processing flow and convert the visual description into visual instructions to present the algorithm processing flow in real time based on the visual instructions.

[0031] The beneficial effects of this invention are:

[0032] 1. This invention utilizes deep integration of generative artificial intelligence to automate the generation of pre-class courseware and dynamically interact with algorithms during class, significantly improving lesson preparation and delivery efficiency, empowering precise teaching, reducing repetitive work, and focusing on the core of instructional design. Using generative AI, the system can automatically generate complete draft courseware containing text lectures, case code, and visualization materials based on the teaching syllabus and knowledge points. Dynamically generating adaptable resources addresses changes in teaching scenarios; the AI ​​generation engine adjusts its needs based on different class learning situations and teaching focuses, dynamically generating personalized teaching cases, practice questions, and accompanying visualizations. It enhances classroom guidance and real-time decision-making capabilities. During in-class demonstrations, teachers no longer need to manually write and debug complex demonstration code. Through the dynamic processing engine, teachers can easily control the demonstration process of algorithms (such as R-CNN and YOLOv3) in real time using natural language commands or parameter sliders. The system provides real-time visual feedback (such as bounding box changes and confidence heatmaps) that allows teachers to intuitively explain algorithm principles and adjust the difficulty and pace of demonstrations in real time based on students' classroom responses, improving the flexibility and effectiveness of classroom guidance.

[0033] 2. This invention is based on deep semantic analysis of teaching syllabi and multimodal teaching resources (text, code, graphs), utilizing Large Language Models (LLMs) and multimodal generative models to automatically generate structurally complete and accurate initial drafts of courseware. Using AI as an intelligent assistant, it provides rich teaching strategy suggestions, multimodal resources (such as charts, cases, code snippets), and generation options while ensuring the accuracy of subject knowledge, ultimately achieving teacher-led quality and efficiency improvement. The depth, breadth, and cases of the generated content can be dynamically adjusted according to different teaching targets (such as undergraduates and graduate students) and class learning conditions. Classroom interaction points or teaching difficulties can be pre-set during the generation process, making pre-class preparation more targeted and flexible. Even teachers without strong programming skills can quickly acquire and integrate interactive materials (such as dynamic web page code) or visualization components for demonstrating algorithms through natural language commands, laying a resource foundation for subsequent dynamic classroom demonstrations.

[0034] 3. This invention integrates code generation and parsing engines, algorithm logic and mathematical principle interpreters, and other technical capabilities to dynamically and visually demonstrate key steps of algorithms such as object detection—including input, feature extraction, candidate box generation, classification and regression, and NMS—in real-time during class, using methods such as feature map changes and detection box generation and adjustment. Students are allowed to adjust model hyperparameters (such as confidence threshold and IoU threshold) through an interactive interface during the demonstration, observing in real-time how parameter changes affect intermediate results and the final detection box output, thus effectively helping students understand the role of core algorithm parameters and optimization strategies. Based on GAI's error diagnosis and debugging assistant and intelligent tutoring capabilities, it provides immediate error location, code optimization suggestions, and concept explanations when students are doing in-class exercises or code implementation, achieving personalized tutoring within the classroom.

[0035] 4. This invention uses data as a link to create a self-feedback and iterative intelligent ecosystem for the entire teaching cycle. Pre-generated courseware and preset interaction points will serve as knowledge points, which will be stimulated and utilized during class. Student interaction data, code debugging performance, and problem feedback generated during class will be recorded by the data acquisition system, realizing a complete link of "pre-setting-generation-interaction-evaluation-optimization". By analyzing the learning effect data during and after class, teaching reflection and optimization suggestions are provided for the pre-generated model or retrieval enhancement generation and teaching strategies, forming an enhanced cycle of "application-feedback-optimization". This enables teaching services to dynamically adapt to the current teaching situation and form a data-based continuous optimization mechanism. Attached Figure Description

[0036] Figure 1 This is a schematic diagram illustrating the dynamic demonstration process of the advanced course algorithm based on generative AI in the embodiment.

[0037] Figure 2 This is a schematic diagram of the dynamic demonstration teaching closed-loop architecture of the course algorithm in the embodiment;

[0038] Figure 3 This is a schematic diagram of the AI ​​generation engine workflow in the embodiment;

[0039] Figure 4 This is a schematic diagram of the data processing flow in the embodiment;

[0040] Figure 5 This is a schematic diagram of data closed-loop driven optimization in the embodiment. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the invention will be further described in detail below with reference to the accompanying drawings and technical solutions.

[0042] Currently, in the field of education, especially in digital and intelligent courses such as "Image Intelligent Processing and Recognition," three major bottlenecks exist: First, the quality of pre-class courseware generation relies on manual secondary processing, lacking personalization and flexibility to adapt to teaching scenarios; second, in-class algorithm demonstrations are static, lacking dynamic, immersive, and interactive presentations of the entire processing process of complex algorithms (such as R-CNN and YOLO); and third, the pre-class, in-class, and post-class teaching stages are technically fragmented, failing to form a complete teaching loop based on data-driven and continuous optimization. Therefore, this invention provides an embodiment, see [link to embodiment]. Figure 1 As shown, a dynamic demonstration method for advanced course algorithms based on generative AI is provided, including:

[0043] S101. Obtain the target algorithm, algorithm parameters, and algorithm input data for the advanced course, and generate teaching courseware for the target algorithm;

[0044] S102. Receive algorithm parameter adjustment data from the teaching interaction of the target algorithm and encode the algorithm parameter adjustment data into structured data; input the structured data and algorithm input data into the pre-trained generative AI model, use the generative AI model to dynamically generate a visual description of the target algorithm processing flow, and convert the visual description into visual instructions to present the algorithm processing flow in real time based on the visual instructions.

[0045] It adopts a real-time processing flow of "data input → data structuring processing → generative large model inference → visualization engine processing → dynamic visualization results". Based on deep fusion generative artificial intelligence (GAI) technology, it dynamically converts input parameters into algorithm processing flow descriptions, enabling dynamic demonstrations of course algorithms throughout the entire teaching cycle. By generating visualization results of algorithm processing flow in real time, it solves the problems of static algorithm display and lack of immersive interactive experience in existing technologies. Through a systematic and integrated technical solution, it fundamentally solves the above pain points and reshapes the teaching paradigm of advanced courses such as "Image Intelligent Processing and Recognition".

[0046] This system utilizes a generative large model to dynamically visualize the algorithm processing flow. Its input data can include algorithm parameters (such as confidence thresholds and NMS thresholds), input image data, and algorithm type (R-CNN, YOLOv3, etc.). During data processing, it receives user-adjusted parameter values, encodes these values ​​as structured input, and converts the input image data into a format that the algorithm can process. A pre-trained generative large model is invoked, passing the structured input to the model. Based on the input parameters and algorithm type, the generative large model dynamically generates a visual description of the algorithm processing flow. The model's output description is converted into visualization instructions, and the system calls a visualization engine to generate real-time dynamic visualization results. These real-time dynamic visualization results (such as bounding box animations, confidence heatmaps, and feature map changes) are presented to students in real time, while simultaneously recording user interaction data. A direct causal relationship exists between the input (algorithm parameters, input image) and output (dynamic visualization results) of the generative large model; changes in parameters directly lead to changes in the algorithm's processing results, and the generative large model is responsible for presenting these changes in real-time in a visual form. This data flow creates a closed loop of "input parameters - algorithm processing - visualization results," enabling students to intuitively understand the impact of parameter adjustments on algorithm output.

[0047] The dynamic visualization process of this invention does not require pre-defined knowledge point relationships. Instead, it visualizes the algorithm processing flow in real time through a generative large model, achieving a closed loop of "dynamic-interactive-feedback". Most existing visualization maps are based on static relationships generated from historical data mining, while the visualization in this invention is based on a dynamic process generated through real-time interaction.

[0048] like Figure 2 The generative artificial intelligence-enabled algorithm dynamic demonstration of the "Image Intelligent Processing and Recognition" course demonstrates how GAI's dynamic generation and understanding capabilities are used to automatically and personalizedly prepare high-quality teaching resources before class, and to visualize and interactively demonstrate the algorithm processing flow during class. This covers the teaching process of "pre-class resource generation - in-class dynamic interaction - post-class feedback optimization". Through data collection and analysis throughout the process, a feedback loop that drives continuous optimization of teaching is formed.

[0049] The system employs several intelligent generation techniques before and after class. First, it generates structured courseware and supporting resources based on the teacher's input of the syllabus, core knowledge points, and optional historical teaching data. Second, during the dynamic interaction phase, teachers and students engage in deep interaction within the classroom environment, using the pre-generated courseware framework and a dynamic processing engine. This engine transforms the static code and flowcharts of deep learning algorithms such as R-CNN and YOLOv3 into a dynamic process with real-time control and visual feedback. Third, in the post-class evaluation and optimization phase, it collects interactive data, practice results, and feedback information generated during class, performs intelligent analysis, and generates learning progress diagnostic reports and teaching reflection suggestions. This data is fed back into the system's AI model and knowledge base to optimize subsequent resource generation strategies and interaction designs, thereby achieving a spiral improvement in teaching effectiveness.

[0050] The teaching materials for the target generation algorithm can be designed to include:

[0051] Based on the needs of algorithm teaching, algorithm teaching courseware is generated using a generative AI model. Several interactive points are set in the algorithm teaching courseware to facilitate teaching interaction. These interactive points are used to visualize the algorithm-related processing flow by adjusting algorithm parameters during algorithm teaching interaction.

[0052] The pre-generated courseware content pre-plans interactive nodes during class (such as parameter adjustment sliders and case switching buttons); the interactive data during class then becomes an important basis for post-class analysis and personalized customization of resources for the next round of class. Thus, teaching is no longer a fragmented process, but an organic, self-evolving whole.

[0053] Specifically, algorithm teaching materials, generated using generative AI models based on algorithm teaching needs, can be designed to include:

[0054] Obtain the algorithm teaching requirements input by the user, which include the teaching objects, teaching objectives, and reference resources;

[0055] By performing multimodal semantic analysis on reference resources, key knowledge units for algorithm teaching are extracted, and a structured knowledge graph for algorithm teaching is constructed.

[0056] The system uses a structured knowledge graph of teaching objects, teaching objectives, and algorithm teaching. It leverages prompting engineering to guide the generation of algorithm teaching courseware using a large language model. Based on the algorithm teaching objectives, it generates algorithm teaching instructions and calls a code generator to generate algorithm code and code examples. Furthermore, it employs a multimodal resource model built on generative artificial intelligence to acquire multimodal teaching resources. These multimodal teaching resources include text, images, videos, and audio corresponding to the algorithm teaching courseware. The multimodal resource model includes a text generation module, an image generation model, and an audio / video generation model.

[0057] The teaching materials for large language model generation algorithms guided by prompting engineering may include:

[0058] Based on the teaching objects and teaching objectives, set up algorithm-structured courseware task descriptions and generate prompts and instructions;

[0059] Interact with the large language model and algorithm teaching structured knowledge graph according to the prompts to obtain structured courseware content;

[0060] Algorithm-based teaching materials are generated from structured courseware content.

[0061] This includes interacting with the large language model and the structured knowledge graph for algorithm teaching based on prompts, which may include:

[0062] Knowledge representation learning is performed on entities and relationships in the structured knowledge graph for algorithm teaching, and the knowledge node vectors are stored in the knowledge base.

[0063] The prompt instructions are converted into query vectors, and the similarity between the query vectors and knowledge node vectors is calculated in the knowledge base to obtain the entities most relevant to the query vectors.

[0064] Using the obtained entities as nodes, perform a depth-first traversal in the graph to find the relationship paths that meet the prompt instructions, and obtain the knowledge subgraph and related reference resource fragments.

[0065] The knowledge subgraph and related reference resource fragments are spliced ​​together and a prompt instruction template is injected to enable the large language model to generate structured courseware content based on the enhanced context.

[0066] As shown in Figure 2, the pre-class AI generation engine is responsible for transforming teachers' teaching intentions into high-quality, interactive pre-class materials. Its implementation logic strictly follows a structured instruction interaction and content generation path, with the core process including three main stages: requirements analysis, content generation, and verification and optimization. Its implementation logic can be summarized as follows:

[0067] 1. Structured Input and Needs Analysis of Teaching Resources. The core elements of the course input by teachers are structured and processed into system-understandable instructions. Input content includes:

[0068] Basic teaching metadata: course name (e.g., "Intelligent Image Processing and Recognition"), target audience, class hours, and teaching objectives (three-dimensional objectives of knowledge, skills, and literacy).

[0069] Core knowledge content: syllabus text, electronic chapters of designated textbooks, and a list of key knowledge points (such as "Fundamentals of Convolutional Neural Networks" and "Principles of the R-CNN Object Detection Algorithm").

[0070] Constraints and Style of Generation: Teachers can specify the level of detail in the generated content, whether to incorporate ideological and political elements into the curriculum, and preferred teaching case areas (such as medical imaging and autonomous driving).

[0071] 2. Knowledge Extraction and Knowledge Graph Construction. The engine utilizes an integrated Large Language Model (LLM) to perform deep semantic analysis on the input text resources. Through knowledge unit extraction and ontology modeling techniques, it identifies and structures the core concepts, algorithm steps, mathematical formulas, and their logical relationships (such as prerequisites, is-a, part-of) in the course, ultimately automatically constructing a structured knowledge graph for the course. This graph serves as the logical foundation for generating all subsequent content.

[0072] 3. Multimodal teaching content generation. Based on the constructed knowledge graph and instructional design theoretical model, the engine calls on different AI capability modules to collaboratively generate content:

[0073] Courseware content and lecture script generation: Based on the lesson plan, automatically generate the outline, page titles, and core explanatory text for PPT courseware. Advanced features can generate a verbatim lecture script for teachers that includes contextual information.

[0074] Code examples and visualization generation: Based on the knowledge point description, the engine calls code generation models (such as CodeLLaMA, DeepSeek-Coder) or links with external computing engines through tool learning mechanisms to generate runnable code examples for core algorithms. For example, for "image filtering," it generates Python code and comments demonstrating the effects of mean filtering and Gaussian filtering. Simultaneously, using a multimodal generation model, it creates visualizations such as diagrams, algorithm flowcharts, and effect comparison charts for abstract concepts.

[0075] 4. Content Synthesis and Human Verification: The generated text, code, images, and other elements are automatically formatted and synthesized to form a preliminary intelligent enhanced courseware package (such as a PPT file with an accompanying code file package). However, according to the data, the generated content is not delivered entirely automatically; a crucial human verification step is necessary. Teachers need to double-verify the generated content, especially highly specialized algorithm principles, mathematical formulas, and code logic, to prevent the illusion of AI. Teachers can use their own teaching experience to further process, optimize, and stylize the content, addressing issues such as unclear emphasis on key points and poor matching between text and images, to create the final usable pre-class materials.

[0076] like Figure 3As shown, the system performs structured processing and knowledge extraction on the multi-source teaching resources provided by teachers, and constructs a machine-understandable course knowledge graph. Subsequently, based on the knowledge graph and instructional design logic, the system concurrently utilizes various generation capabilities (such as long text generation, code generation, and multimodal generation) to produce lesson plan text, accompanying code examples, visualization materials, and pre-set classroom interaction points in one go. Finally, all generated content is automatically integrated and formatted to form a complete courseware resource package for teachers to verify.

[0077] This engine provides a teacher interaction interface, an internal model scheduling interface, and a data output interface. The teacher interaction interface offers a web-based graphical console and a natural language dialogue interface, supporting multiple input methods such as file upload, form filling, and dialogue commands. The internal model scheduling interface features a unified model API gateway that dynamically schedules and combines different underlying AI models (such as LLM, code models, and multimodal models) based on the type of generated task (text, code, image). The data output interface outputs generated standard lesson plans (Word / PDF), courseware (PPT), code packages (.zip), and associated resource links through a standard file service interface, and can be pushed to the in-class interaction platform.

[0078] Taking a teaching unit in the course "Intelligent Image Processing and Recognition" as an example, the process of automatically generating pre-class courseware can be summarized as follows:

[0079] (1) Teacher input

[0080] In the AI ​​generation engine console, teachers select the course "Intelligent Image Processing and Recognition," and then choose the sub-module "Object Detection." They input the core knowledge points: "Principles and Process of the Two-Stage Detection Algorithm R-CNN," "Core Ideas of the Single-Stage Detection Algorithm YOLOv3," and "Post-processing Concepts such as Intersection over Union (IoU) and Non-Maximum Suppression (NMS)." They also upload the corresponding PDF chapters of the course textbook. In "Case Preferences," they select "Traffic Scenarios."

[0081] (2) Data processing and generation

[0082] Knowledge graph construction: AI analyzes the textbook, extracts concepts such as "object detection", "R-CNN", "Selective Search", "CNN feature extraction", "SVM classification", "boundary regression", "YOLOv3", "mesh generation", "anchor", "IoU", and "NMS", and establishes connections (e.g., R-CNN includes Selective Search, and NMS is used to eliminate redundant boxes).

[0083] Lesson plan generation: Based on the graph, a detailed lesson plan is generated. For example, the system automatically designs a teaching activity that "guides students to understand the trade-off between 'accuracy vs. speed' by comparing the flowcharts of R-CNN and YOLOv3".

[0084] (3) Courseware and resource generation.

[0085] PPT Content: Generate a PPT page that includes an algorithm comparison table and a step-by-step flowchart (generated from a multimodal model).

[0086] Code examples: Generate two runnable Python scripts: demo_rcnn.py (demonstrating the R-CNN workflow using a simplified model) and demo_yolov3.py (calling a PyTorch Hub pre-trained model). The code includes comments for key steps and modifiable parameter variables (such as conf_threshold).

[0087] Visualization materials: Generate a set of traffic images containing vehicles and pedestrians as example input. Generate a dynamic diagram of the IoU calculation process (such as an animation of the area change of two rectangles from separation to overlap).

[0088] Interactive presets: In the generated PPT, mark the two parameters conf_threshold and nms_threshold as interactive points, and indicate in the notes that "This can be connected to the dynamic processing engine for real-time demonstration".

[0089] (4) Teacher verification and finalization: Teachers review the generated content and correct any deviations in the AI's understanding.

[0090] The dynamic processing engine transforms deep learning algorithm teaching from static explanation to an immersive dynamic experience. It leverages the code understanding, real-time reasoning, and interactive response capabilities of generative AI to construct an algorithm demonstration and experimental environment. Based on the algorithm teaching materials, the algorithm demonstration screen is divided into a code area, a parameter control area, and a visualization area. In the algorithm interaction, the code area uses an AI agent to display the algorithm code; the parameter control area uses an AI agent to display algorithm parameters and allow interaction through parameter adjustment; and the visualization area uses an AI agent to visualize the algorithm processing during the interaction. Furthermore, it can generate a learning profile of the algorithm based on interaction data and obtain algorithm teaching optimization suggestions. These suggestions are fed back to the generative AI model, which then optimizes the algorithm teaching materials using a combination of cloud inference and edge computing. The implementation logic of this engine is summarized as follows:

[0091] 1. Dynamic code generation and logic parsing

[0092] When explaining algorithms such as R-CNN and YOLOv3, teachers or students can input natural language commands (e.g., "Please demonstrate the detection process of YOLOv3 on this traffic image") in the interactive interface. The engine calls the code generation and parsing module, which not only generates executable object detection code, but more importantly, can explain the algorithm logic step by step and interactively. For example, when demonstrating R-CNN, the intermediate results of "Region Proposal -> Feature Extraction -> Classification & BBox Regression" can be shown step by step.

[0093] 2. Real-time parameter adjustment and visual feedback

[0094] Providing graphical controls (such as sliders and drop-down menus) allows students and teachers to adjust algorithm parameters in real time. This is key to achieving dynamic processing. Taking object detection as an example, adjustable parameters include: Model Confidence Threshold: Observing changes in the number of detected boxes in real time; Non-Maximum Suppression (NMS) Threshold: Intuitively understanding the merging process of overlapping boxes; Input Image Size: Observing the differences in processing speed and accuracy of the model for inputs of different resolutions.

[0095] Each time a parameter is adjusted, the engine will re-execute the relevant code snippet or call the inference interface of the loaded model in real time, and immediately display the processed results on the screen through visualization charts, highlighting, and other methods.

[0096] Changes in input parameters (such as adjusting the confidence threshold from 0.5 to 0.2) directly lead to changes in the algorithm's processing results (the number of detection boxes increases from 25 to 45). The generative large model is responsible for presenting these changes in a real-time visual form, enabling students to intuitively understand the impact of parameter adjustments on the algorithm's output. The dynamic visualization process is real-time and interactive, allowing students to adjust parameters instantly and see the changes in results, forming a closed loop of "input-processing-output".

[0097] 3. Real-time error diagnosis and collaborative optimization

[0098] As students practice writing or modifying code during class, the engine's error diagnosis module analyzes the code in real time, locating syntax and logic errors and providing suggestions for improvement. Furthermore, it can guide students in algorithm optimization. For example, after a student implements basic object detection code, it might prompt, "The current NMS implementation is inefficient; what methods can be used to optimize it?" and guide the student to think about or directly provide a comparison of optimized code versions.

[0099] 4. Immersive Visualization and Scene Construction:

[0100] Beyond traditional code and diagrams, the engine leverages the descriptive capabilities of generative AI to drive advanced visualizations. For example, for feature map visualization of convolutional neural networks, AI can generate and control a dynamic, layer-by-layer activation visualization animation. For courses such as 3D point cloud processing, dynamic 3D models can be generated for interactive demonstrations.

[0101] This engine provides interactive classroom interfaces, computational backend interfaces, and data acquisition interfaces to enable dynamic visualization. The interactive classroom interface, integrated into the interactive screen of a smart classroom or a student's personal learning terminal, offers multiple interactive components such as a code editor, parameter control panel, visualization view area, and instant chat (for interacting with AI teaching assistants). The computational backend interface interfaces with cloud or local GPU computing resource pools, providing lightweight, low-latency model inference services. For complex models (such as YOLOv3), a pre-loaded model and real-time inference using API parameters are typically employed. The data acquisition interface records all interactive events in real time, including parameter adjustment sequences, code submission versions, Q&A records, practice time, and results. This data is a crucial source for constructing the teaching loop.

[0102] In the teaching unit example of the course "Intelligent Image Processing and Recognition," the instructor first explains the core idea of ​​YOLOv3: dividing an image into a grid, predicting bounding boxes and categories for each grid. By clicking the interactive points marked on the PPT, the system seamlessly switches to the interface of the dynamic processing engine. The left side of the interface displays the code demo_yolov3.py generated before class, and the right side is the visualization area. The dynamic visualization process is described as follows: The preprocessed input image is automatically loaded; the generative large model receives the algorithm type (YOLOv3) and default parameters (confidence threshold of 0.5) as input; the model dynamically generates the algorithm processing flow description: "At a confidence threshold of 0.5, the model outputs 25 detection boxes, of which 10 boxes have a confidence level below the threshold and will be filtered out." The visualization engine generates a dynamic visualization based on this description, displaying 25 detection boxes on the image, with 10 boxes marked as low confidence; the student adjusts the confidence threshold to 0.2, and the system automatically triggers the generative large model to re-infer; the large model generates a new description: "At a confidence threshold of 0.2, the model outputs 45 detection boxes, of which 25 boxes have a confidence level below the threshold and will be filtered out." The visualization engine updates the display, showing 45 detection boxes on the image, with 25 boxes marked as low confidence. The entire process requires no manual intervention from the teacher; the system automatically completes the conversion from parameter adjustment to visualization results.

[0103] like Figure 4As shown, the data input area contains three types of input sources: real-time adjustable algorithm parameters (confidence threshold 0.5, NMS threshold 0.45, image size 416×416), supporting input images in formats such as PNG / JPEG, and algorithm types obtained from courseware (such as YOLOv3, R-CNN, etc.); the data processing area contains three modules: "Parameter Encoding and Data Preprocessing" automatically processes the input (without manual intervention), and the output points to "Generative Large Model (CodeLLaMA / DeepSeek-Coder)" to generate a real-time description of the algorithm processing. The model output is then passed to the "Visualization Engine" for real-time rendering (without pre-generation); the data output area is divided into two parts: "Dynamic Visualization Results" (such as bounding box animation, confidence heatmap, feature map change process, updated in real time with parameter changes) are presented to students in real time through data feedback (purple, dashed line); "Interactive Data" (such as parameter adjustment sequence, interaction duration, question submission, automatically collected) is also automatically collected through data feedback. In the process, solid lines represent the main data flow, and dashed lines represent the auxiliary data flow (the auxiliary path from output to feedback), clearly distinguishing the core and auxiliary logic of data flow, while color is used to enhance the classification of functional modules.

[0104] A classroom data report is automatically generated after class, recording the numerical ranges most students attempted for `conf_threshold` and `nms_threshold`, as well as the types of code errors made during practice. AI analysis of student actions identifies areas of comprehension bias. For these biased areas, detailed comments and common error examples are added to the code examples generated before the next class. These results are directly fed back to teachers to adjust subsequent teaching strategies; they are also integrated into the teaching system's data platform. When teachers prepare a "target detection" lesson for a similar class next time, the AI ​​generation engine will refer to this historical data to generate more precise lesson plan wording emphasizing threshold correlations and more targeted error-prevention code comments, thus initiating a new and optimized teaching cycle.

[0105] like Figure 5 As shown, fine-grained teaching data (such as pre-class preparation, classroom parameter adjustment records, Q&A content, and homework / test results) is continuously collected throughout the entire teaching cycle—before, during, and after class. This data is aggregated in the data backend for analysis, generating learning profiles for classes and individuals, and diagnosing teaching difficulties and common errors. Specific optimization suggestions generated based on this analysis are directly fed back to the AI ​​generation engine, driving the next round of personalized and precise generation of teaching resources. This allows the system to continuously evolve based on real teaching feedback data, achieving an intelligent teaching ecosystem.

[0106] Furthermore, based on the above method, this embodiment of the invention also provides a dynamic demonstration system for advanced course algorithms based on generative AI, comprising: a courseware generation module and a dynamic interaction module, wherein,

[0107] The courseware generation module is used to obtain the target algorithm, algorithm parameters, and algorithm input data for advanced courses, and generate teaching courseware for the target algorithm.

[0108] The dynamic interaction module is used to receive algorithm parameter adjustment data in the teaching interaction of the target algorithm and encode the algorithm parameter adjustment data into structured data. The structured data and algorithm input data are input into a pre-trained generative AI model. The generative AI model is used to dynamically generate a visual description of the target algorithm processing flow and convert the visual description into visual instructions to present the algorithm processing flow in real time based on the visual instructions.

[0109] The following section uses the core chapter "Object Detection" in the course "Intelligent Image Processing and Recognition" as an example to elaborate on the implementation process of this solution, covering the entire process of pre-class preparation, in-class teaching, and post-class feedback, and discusses key issues and applicability adjustments during implementation.

[0110] Course Title: Intelligent Image Processing and Recognition (Third-year undergraduate, elective course)

[0111] Teaching Chapter: Principles and Applications of Object Detection Algorithms

[0112] Key knowledge points: The core ideas, network architecture, loss functions, and performance comparison of two-stage detection algorithms (taking R-CNN as an example) and single-stage detection algorithms (taking YOLOv3 as an example).

[0113] Course schedule: A total of 4 class hours. Of these, 2 class hours are for theoretical explanation and dynamic demonstration of algorithms, and 2 class hours are for student hands-on practice and project discussions.

[0114] Implementation environment: A smart classroom equipped with an intelligent interactive screen, where students can access the teaching system via personal computers or tablets; the server side is equipped with the teaching system of this invention, which integrates a generative AI engine (calling model APIs such as DeepSeek and GLM) and inference services for lightweight object detection models (YOLOv3-tiny, a simplified version of Faster R-CNN).

[0115] 1. AI-assisted pre-class courseware generation

[0116] The lecturer prepares the lesson using the department's AI-generated engine.

[0117] (1) Teacher input and demand definition

[0118] The teacher inputs the teaching objectives for this lesson into the system: "To enable students to master the basic principles of R-CNN and YOLOv3, and understand their advantages, disadvantages and applicable scenarios."

[0119] Upload the chapter descriptions of "Object Detection" from the course syllabus and scanned copies of relevant pages from the classic textbook "Computer Vision: Algorithms and Applications" as reference resources.

[0120] Input natural language command: "Generate a detailed lesson plan that includes learning analysis, teaching objectives, key points and difficulties, and teaching process (including introduction, algorithm comparison and explanation, dynamic demonstration design, and classroom exercises). The plan should focus on designing an interactive dynamic demonstration that allows students to adjust parameters in real time to observe the changes in detection performance of the two algorithms in different scenarios (such as occlusion and small targets). Provide a corresponding PPT outline and three code examples ranging from simple to complex."

[0121] (2) Data processing and content generation

[0122] Multimodal semantic parsing technology is employed to analyze input resources such as textbooks and syllabi, extracting key knowledge units such as "selective search," "candidate regions," "bounding box regression," "non-maximum suppression (NMS)," and "grid prediction." A structured knowledge graph for this chapter is then constructed to clarify the relationships between concepts. Based on knowledge graphs and instructional design theory, multiple generative AI models are invoked in parallel, as summarized below:

[0123] Lecture Notes Generation: A complete lesson plan draft for this class is generated by a large language model, with a clear content structure. For example, in the "Teaching Process", the AI ​​not only lists the explanation steps, but also presets multiple "interactive points in class", such as: "Interactive Point 1: Pause here, start the dynamic demonstration interface, and guide students to compare the intuitive differences in speed and accuracy between R-CNN and YOLOv3 when processing the same image containing multi-scale objects."

[0124] PPT presentation generation: Based on the lesson plan content, automatically generate a draft PPT slide, including algorithm flowcharts, core formulas, comparison tables, etc., and reserve an interface for dynamic demonstration plugins.

[0125] The code and visualization materials are generated, including three gradient-style code examples: Example 1 is a complete script for loading a pre-trained YOLOv3 model in PyTorch and performing single-image prediction; Example 2 shows how to modify the confidence threshold and NMS threshold; and Example 3 provides a simplified Faster R-CNN Region Proposal Network (RPN) implementation framework for students with extra learning capacity to explore.

[0126] As instructed, the AI ​​generation engine will also utilize its tool learning capabilities to generate several dedicated visual material prompts for this lesson, such as: "Generate a city street scene image containing dense pedestrians, small vehicles in the distance, and partially occluded objects for comparison and demonstration of the object detection algorithm's performance."

[0127] (3) Teacher manual verification and finalization

[0128] After receiving the results generated by the system, teachers conduct rigorous professional verification. The focus is on checking the accuracy of the AI's description of the algorithm principles, the objectivity of the comparative analysis, and the absence of logical errors in the generated code. Authoritative textbooks and literature can be consulted, and Retrieval-Enhanced Generation (RAG) technology can be integrated to connect the AI ​​model to the course knowledge base, thereby improving the accuracy of the answers.

[0129] Teachers should personalize their lesson plans based on their own experience, and need to increase the amount of explanation on "non-maximum suppression," a common difficulty for students; and further refine the prompts in the AI-generated street view images, requiring the addition of rain and fog effects to enhance the robustness of the algorithm.

[0130] Once approved, the lesson plan is finalized, and the marked "interaction points" along with associated code examples and visual materials will be synchronized to the in-class dynamic processing engine, making it ready.

[0131] 2. Dynamic interaction and algorithm demonstration during class

[0132] The teacher uses the finalized courseware and activates the dynamic processing engine at preset "interaction points" to achieve real-time, interactive teaching of the algorithm. After explaining the basic process of R-CNN to the "generating candidate regions" step according to the PPT, the teacher clicks the "start demonstration" button on the courseware, and the system interface switches to dynamic demonstration mode.

[0133] The dynamic processing engine is in operation and interacting with teachers and students. At this time, the screen is divided into the following three areas:

[0134] Code section: Displays the core function code of YOLOv3 detection in real time, with key parameters (such as conf_threshold and nms_threshold) highlighted.

[0135] Parameter control area: Provides sliders and drop-down menus, which can adjust the confidence threshold and NMS threshold in real time, and can select different input images (such as "sunny street scene" and "rainy fog street scene" generated before class).

[0136] Visualization Area: The main display area, rendering the algorithm processing in real time. For example, it shows the YOLOv3 grid division step by step, the predicted bounding box for each grid, the bounding box after threshold filtering, and the final result after NMS processing. All bounding boxes and confidence scores are displayed dynamically.

[0137] At this point, students can be guided to adjust the parameters, dragging the confidence threshold from 0.5 to 0.8, and observe the changes in the visualized area. After the student's operation, the system immediately re-inferences, and many low-confidence detection boxes disappear from the visualized area. The teacher can then use this opportunity to explain the physical meaning of confidence and its trade-off between precision and recall.

[0138] The teacher switched to R-CNN demonstration mode and selected the "Rainy Fog Street View" image. Students observed that R-CNN's detection results showed more missed detections. The teacher prompted students to think: "Compared to the results of YOLOv3, why might the two-stage algorithm perform worse here? Which stage might it be related to?" Students proposed hypotheses through the AI ​​dialogue area and discussed them with the teacher and AI teaching assistant.

[0139] 3. Post-class data closed loop and teaching optimization

[0140] (1) Full-cycle data collection. The background silently collects data throughout the entire process, including: the duration and dwell time of students' pre-lesson preparation, classroom interaction logs (the number of times and values ​​each student adjusted the slider, the content of questions asked in the AI ​​dialogue area), the history of code submissions and error types for computer practice, and the completion status and accuracy of online exercises after class.

[0141] (2) Generation of learning situation analysis and diagnostic report:

[0142] After class, the system automatically analyzes the collected data and generates a class learning progress diagnostic report. Simultaneously, personalized learning suggestions are generated and sent to each student.

[0143] (3) Feedback and Iterative Optimization:

[0144] After reviewing the learning progress report, teachers identified key difficulties that needed to be addressed in the next round of teaching. This report, along with students' personalized feedback data, was automatically fed back to the AI ​​generation engine via the API.

[0145] When teachers prepare lessons on the same chapter for parallel classes, the AI ​​generation engine will prioritize adding more illustrated and animated examples of NMS when generating lesson plans and courseware. It will also enhance the annotations for bounding box coordinate encoding and provide common error prevention tips in the generated code examples. A combined cloud inference and edge computing model can be adopted. Lightweight model interaction requests are processed locally or on edge servers; complex retraining or large-scale re-inference tasks are queued by the system and submitted to cloud GPU clusters such as big data centers for asynchronous processing, ensuring smooth classroom interaction. This achieves a complete teaching loop based on real teaching data: "pre-setting-generation-interaction-evaluation-optimization".

[0146] To address the issue of students over-relying on AI and neglecting fundamental programming skills, periodic tests or core assignments can require students to handwrite key algorithm steps or complete code completion tasks in an isolated environment. The teaching philosophy emphasizes that AI is a critical thinking partner and an efficiency tool, not an answer generator. Furthermore, to address the varying levels of student ability leading to different interaction paces, AI can generate pre-class materials of varying difficulty for students to choose from. During class, a dynamic demonstration system can provide a basic mode and an inquiry mode; the former focuses on the intuitive effects of parameter adjustments, while the latter provides visualization of deeper network intermediate layer features.

[0147] For different teaching scales, such as small group teaching / seminar classes, multi-agent collaborative inquiry scenarios can be added. For example, students can be divided into groups and take on the roles of parameter tuners, result observers, and algorithm questioners, interacting and debating with the AI ​​roles in the system to deeply deconstruct the algorithm. For large-scale open courses, the focus is on generating high-quality resources before class and large-scale learning analysis after class. In-class interaction is mainly teacher-led, all-inclusive presentations, combined with lightweight interactions such as voting and bullet comments. The dynamic processing engine primarily serves the teacher's presentation preparation.

[0148] For different learning groups, such as beginners (e.g., lower-level undergraduates), AI-generated courseware can be prompted with suggestions like "reduce theoretical depth, increase intuitive analogies and animated GIFs." In-class dynamic demonstrations should reduce adjustable parameters, highlighting the 1-2 most crucial parameters (e.g., confidence level), prioritizing visualization rendering speed, and emphasizing the observability of results. For students with prior knowledge (e.g., graduate students or senior undergraduates), AI-generated content can include introductions to cutting-edge papers, comparisons of algorithm variants (e.g., YOLOv5, v7), and case studies of applications in specific fields (medical imaging, autonomous driving). In-class interaction can provide more visualization of underlying parameters and internal model features, encouraging students to optimize model performance on specific datasets by adjusting parameters, thus bridging research training.

[0149] As seen in the specific examples above, this solution systematically integrates generative AI and dynamic interactive technology into the entire teaching process of digital intelligence courses such as "Intelligent Image Processing and Recognition." By using a generative large-scale model to dynamically generate a visual description of the algorithm's processing flow, students can adjust parameters in real time and observe changes in the algorithm's internal processing state. This allows them to establish a systematic and visual understanding of the algorithm's working principles, not only significantly improving lesson preparation and delivery efficiency but, more importantly, deepening students' understanding of complex principles through dynamic, visual, and interactive algorithm demonstrations. Changes in input parameters (such as confidence thresholds) directly drive the generative large-scale model to dynamically generate a description of the algorithm's processing flow, which in turn generates dynamic results in real time through a visualization engine, forming a closed loop of "input parameters → algorithm processing → visualization results." Through this data-driven closed loop, continuous adaptive optimization of the teaching content is achieved, fully demonstrating its practical value and innovation in promoting the digital transformation of higher education teaching models.

[0150] This invention transforms the abstract into the concrete, making the algorithm's principles visible and interactive. It transforms deep learning algorithms from static code, formulas, or flowcharts into a fully interactive, dynamic processing flow. Students can witness firsthand every key step from inputting an image, through feature extraction, region nomination, classification and regression, to the final output detection box. By adjusting parameters in real time (such as IoU threshold and confidence threshold), students can immediately see the chain reaction of parameter changes on the detection results, thereby gaining a deeper understanding of the algorithm's internal working mechanism and the significance of hyperparameters, breaking through the "black box" cognitive bottleneck. It also provides an immersive and exploratory practical environment. The system's interactive demonstration environment acts as a virtual experimental platform. Students can safely conduct "hypothesis-verification" exploration in the classroom, greatly stimulating their learning interest and cultivating their computational thinking and problem-solving abilities. It provides immediate and personalized learning feedback and guidance. During classroom exercises or after-class practice, when students write or modify algorithm code, the integrated intelligent tutoring module provides real-time error diagnosis, code optimization suggestions, and concept explanations.

[0151] This invention integrates data across the entire teaching lifecycle to achieve precise learning progress diagnosis. Utilizing a unified teaching data platform, it systematically collects data from pre-class resource generation and in-class interactive operations to post-class practice feedback. Through data analysis, the system automatically generates learning progress diagnostic reports for classes and individual students, accurately pinpointing knowledge gaps (such as common misunderstandings among students regarding Non-Maximum Suppression (NMS) parameters), providing teachers with objective criteria for evaluating teaching effectiveness. Driving continuous self-optimization of teaching content, learning progress data and feedback can be fed back into the pre-class AI generation engine. In subsequent courses on the same topic, this automatically expands the explanation of that knowledge point, adding more targeted visual examples or interactive demonstrations.

[0152] In conclusion, this solution not only significantly improves the teaching efficiency and intuitiveness of digital courses such as "Intelligent Image Processing and Recognition" at the practical level, but also constructs a self-evolving, data-driven intelligent teaching ecosystem at the system level, which has significant teaching practice value and prospects for promotion and application.

[0153] Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention.

[0154] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0155] The units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations are not considered to be beyond the scope of this invention.

[0156] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiments can be implemented in hardware or as a software functional module. This invention is not limited to any particular combination of hardware and software.

[0157] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic demonstration method for advanced course algorithms based on generative AI, characterized in that, Include: Obtain the target algorithm, algorithm parameters, and algorithm input data for advanced courses, and generate teaching courseware for the target algorithm; The system receives algorithm parameter adjustment data from interactive teaching of the target algorithm and encodes the algorithm parameter adjustment data into structured data. The structured data and algorithm input data are input into a pre-trained generative AI model. The generative AI model is used to dynamically generate a visual description of the target algorithm processing flow and convert the visual description into visual instructions to present the algorithm processing flow in real time based on the visual instructions.

2. The method for dynamically demonstrating advanced course algorithms based on generative AI according to claim 1, characterized in that, Teaching materials for the target generation algorithm, including: Based on the needs of algorithm teaching, algorithm teaching courseware is generated using a generative AI model. Several interactive points are set in the algorithm teaching courseware to facilitate teaching interaction. These interactive points are used to visualize the algorithm-related processing flow by adjusting algorithm parameters during algorithm teaching interaction.

3. The method for dynamically demonstrating advanced course algorithms based on generative AI according to claim 2, characterized in that, Based on the needs of algorithm teaching, algorithm teaching courseware is generated using generative AI models, including: Obtain the algorithm teaching requirements input by the user, which include the teaching objects, teaching objectives, and reference resources; By performing multimodal semantic analysis on reference resources, key knowledge units for algorithm teaching are extracted, and a structured knowledge graph for algorithm teaching is constructed. The system uses a structured knowledge graph of teaching objects, teaching objectives, and algorithm teaching. It leverages prompting engineering to guide the generation of algorithm teaching courseware using a large language model. Based on the algorithm teaching objectives, it generates algorithm teaching instructions and calls a code generator to generate algorithm code and code examples. Furthermore, it employs a multimodal resource model built on generative artificial intelligence to acquire multimodal teaching resources. These multimodal teaching resources include text, images, videos, and audio corresponding to the algorithm teaching courseware. The multimodal resource model includes a text generation module, an image generation model, and an audio / video generation model.

4. The method for dynamically demonstrating advanced course algorithms based on generative AI according to claim 3, characterized in that, Teaching materials on algorithms for generating large language models using prompting engineering, including: Based on the teaching objects and teaching objectives, set up algorithm-structured courseware task descriptions and generate prompts and instructions; Interact with the large language model and algorithm teaching structured knowledge graph according to the prompts to obtain structured courseware content; Algorithm-based teaching materials are generated from structured courseware content.

5. The method for dynamically demonstrating advanced course algorithms based on generative AI according to claim 4, characterized in that, The system interacts with the large language model and the structured knowledge graph for algorithm teaching, based on the prompts and instructions. This includes: Knowledge representation learning is performed on entities and relationships in the structured knowledge graph for algorithm teaching, and the knowledge node vectors are stored in the knowledge base. The prompt instructions are converted into query vectors, and the similarity between the query vectors and knowledge node vectors is calculated in the knowledge base to obtain the entities most relevant to the query vectors. Using the obtained entities as nodes, perform a depth-first traversal in the graph to find the relationship paths that meet the prompt instructions, and obtain the knowledge subgraph and related reference resource fragments. The knowledge subgraph and related reference resource fragments are spliced ​​together and a prompt instruction template is injected to enable the large language model to generate structured courseware content based on the enhanced context.

6. The method for dynamically demonstrating advanced course algorithms based on generative AI according to claim 1, characterized in that, Based on visualization commands, the algorithm processing flow is presented in real-time and dynamically. It also includes: Based on the algorithm teaching materials, the algorithm demonstration screen is divided into a code area, a parameter control area, and a visualization area. In the algorithm interaction, the algorithm code is displayed in the code area by an AI agent, the algorithm parameters are displayed in the parameter control area by an AI agent and the interaction is achieved by adjusting the algorithm parameters, and the algorithm processing process is visualized in the visualization area by an AI agent.

7. The method for dynamically demonstrating advanced course algorithms based on generative AI according to claim 1 or 6, characterized in that, Also includes: Based on the interactive data, a learning profile of the algorithm is created and suggestions for optimizing algorithm teaching are obtained. These suggestions are then fed back to the generative AI model, which is used to optimize the algorithm teaching materials using a combination of cloud inference and edge computing.

8. A dynamic demonstration system for advanced course algorithms based on generative AI, characterized in that, It includes: a courseware generation module and a dynamic interaction module, among which, The courseware generation module is used to obtain the target algorithm, algorithm parameters, and algorithm input data for advanced courses, and generate teaching courseware for the target algorithm. The dynamic interaction module is used to receive algorithm parameter adjustment data in the teaching interaction of the target algorithm and encode the algorithm parameter adjustment data into structured data. The structured data and algorithm input data are input into a pre-trained generative AI model. The generative AI model is used to dynamically generate a visual description of the target algorithm processing flow and convert the visual description into visual instructions to present the algorithm processing flow in real time based on the visual instructions.

9. An electronic device, characterized in that, include: At least one processor, and a memory coupled to said at least one processor; The memory stores a computer program that can be executed by the at least one processor to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, enables the implementation of the method as described in any one of claims 1 to 7.

Images