Improvement of Model Collapse Phenomenon due to Artificial Intelligence Recursive Learning

Object recognition technology is widely used as a basic technology for process automation, autonomous driving, and smart system construction in the field of artificial intelligence.In such object recognition technology, recursive learning techniques are widely used to improve the accuracy of recognition, in which images generated by artificial intelligence are reused as learning data.However, this recursive learning can cause the problem of model collapse, which reduces the performance of object recognition.To solve this problem, in this paper, we mathematically analyzes the image quality degradation process that causes model collapse, and propose a generated image evaluation technique using singularity analysis and natural logarithm function based on this.In addition, the image evaluation technique in the paper was experimentally implemented, and analysis was performed on images with different damage ratios from the original image.Through these experiments and experimental results, it was shown that the image evaluation technique proposed in the paper can be sufficiently utilized to improve the model collapse phenomenon by early detecting image damage that can occur in model collapse of recursive learning.