Defect detection method and device for deep learning library based on fuzzy test, and medium

The invention relates to the field of deep learning, in particular to a deep learning library defect detection method and device based on fuzzy testing and a medium, which can increase the ability to explore codes of a deep learning library, improve the ability to detect library defects, and improve the detection efficiency by generating a model with high diversity. Specifically, models with rich structures, rich parameter values and rich weight values are generated as much as possible, so that the exploration capability of deep learning library codes is improved; defects of the deep learning library are detected in the model training and prediction stage, and the library defect detection capability is improved. 本发明涉及深度学习领域，具体涉及一种基于模糊测试的深度学习库缺陷检测方法、设备和介质,能够增加对深度学习库代码的探索能力，提高库缺陷检测的能力,通过生成多样性高的模型，具体是尽可能生成结构丰富、参数取值丰富以及权重取值丰富的模型，从而增加对深度学习库代码的探索能力；在模型训练和预测阶段检测深度学习库缺陷，提高库缺陷检测的能力。