A real-time personalized AI teaching ultra-lightweight model design method and system

By using a lightweight Transformer neural network structure and lightweight technology, the problem of poor adaptability of AI teaching systems on multiple terminal devices is solved, realizing low-cost, high-efficiency personalized teaching, which is suitable for resource-constrained environments.

CN122133718APending Publication Date: 2026-06-02TIANJIN ZHIXIN COSLIGHT TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
TIANJIN ZHIXIN COSLIGHT TECHNOLOGY CO LTD
Filing Date
2026-03-25
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing AI teaching systems cannot be embedded and adapted to multiple terminal devices, and they suffer from high prices, high maintenance costs, and low compatibility with teaching materials, making it difficult to achieve personalized teaching.

Method used

Design an ultra-lightweight model for real-time personalized AI teaching. Employ a lightweight Transformer neural network structure and combine attention head pruning, knowledge distillation, and quantization perception techniques to reduce the model's parameter scale and computational complexity, making it suitable for resource-constrained environments.

Benefits of technology

It enables lightweight deployment of the model on edge devices, reduces learning costs, improves teaching quality and learning efficiency, and adapts to the personalized teaching needs of multiple terminal devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122133718A_ABST
    Figure CN122133718A_ABST
Patent Text Reader

Abstract

This invention relates to a lightweight model design method and system for real-time personalized AI teaching. The method involves designing a lightweight large language model as the teacher model, training the teacher model using a training set, and employing structured pruning techniques to remove redundant attention heads and feedforward network layers from the teacher model, generating a streamlined student model. Knowledge distillation is used to transfer the attention patterns and hidden layer representations of the teacher model to the student model. Finally, a mixed-precision quantization method is used to compress and optimize the student model to obtain a lightweight teaching model. This reduces the magnitude of model parameters and computational complexity, making it easy to deploy and run, suitable for resource-constrained environments, and simultaneously providing efficient knowledge transfer and personalized tutoring, thereby improving teaching quality and learning efficiency.
Need to check novelty before this filing date? Find Prior Art