A Method and System for Training AI Video Generation Models Based on Cognitive Neural Data

By acquiring physiological feedback data from video samples, extracting target neural response features, calculating and optimizing the joint loss, and training a video generation model, the problem of insufficient accuracy in emotional expression in existing video generation methods is solved, and the controllability of emotional expression and psychological impact of generated videos is achieved.

CN122390002APending Publication Date: 2026-07-14LIAONING UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
LIAONING UNIVERSITY
Filing Date
2026-04-22
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing video generation methods lack objective quantitative data on the internal cognitive and emotional responses of humans when watching videos, resulting in videos that lack accuracy in emotional expression and controllability in psychological impact.

Method used

By acquiring sample video data, text description data, and physiological feedback data from video samples, the target neural response features of the physiological feedback data are extracted, a mapping relationship is constructed, and the joint loss is calculated and optimized to train the video generation model to improve the accuracy of emotional expression.

Benefits of technology

The video generation model not only learns pixel-level reconstruction, but also learns to generate video content that conforms to expected emotions and attention patterns, thereby improving the accuracy of emotional expression in generated videos and the controllability of the psychological impact on the audience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122390002A_ABST
    Figure CN122390002A_ABST
Patent Text Reader

Abstract

The application relates to an AI video generation model training method and system based on cognitive neural data. The method comprises the following steps: acquiring sample video data, text description data and physiological feedback data of a video sample; performing feature extraction on the physiological feedback data to obtain target neural response feature data; constructing a mapping relationship between the sample video data and the target neural response feature data to obtain a neural response prediction model; inputting the text description data and predicted video data into a preset initial video generation model to obtain predicted video data and predicted neural response feature data; calculating an optimization joint loss based on the predicted video data, the sample video data, the predicted neural response feature data and the target neural response feature data; and performing model optimization processing on the initial video generation model in the direction of reducing the optimization joint loss to obtain a trained video generation model. The method can improve the emotional expression accuracy of the generated video.
Need to check novelty before this filing date? Find Prior Art