Deep Learning Model Risk Identification and Control Technology

Haoran Zhang, Guang Yang, Hongyun Yu, Lingzhong Meng, Baolin Yan

Abstract: To reduce the probability of hidden dangers introduced by deep learning models and enhance their security and reliability, this paper takes the data and algorithms of deep learning models as the core objects and proposes a risk identification and control method based on the lifecycle of deep learning models. This paper first combs the activities and contents in the whole lifecycle of deep learning models and constructs a deep learning lifecycle model. Then, based on the lifecycle model, it analyzes the risk factors existing in each development stage of deep learning and identifies the weak links that may introduce security risks. Finally, for the risk factors at each stage, possible control measures and solutions are proposed to provide necessary decision support for the improvement and quality enhancement of deep learning models.

External IDs:dblp:conf/dsa/ZhangYYMY24