深度学习智能软件的可靠性测试研究综述OA

Due to the inherent vulnerabilities and the lack of interpretability of deep learning networks,the intelligent sys-tems that they support face operational reliability risks.Consequently,the reliability of deep learning-based intelligent software has become a key focus and challenge in the field of software testing.However,systematic reviews specifically addressing the reliability testing of deep learning-based intelligent software remain relatively scarce.To address this gap,this paper analyzes the uncertainty factors affecting the reliability of deep learning-based intelligent software from three dimensions:data,models(algorithms),and platforms(frameworks).Following the standard workflow of reliability-oriented software testing,the paper systematically reviews the major challenges encountered across four core stages:test case generation,test method selection,test execution,and result evaluation,along with the key technical advances in related areas.On this basis,several future research directions are outlined,including determining performance boundaries of intel-ligent software,designing lightweight testing methods,overcoming adversarial example predicament,improving model robustness,reliability testing of large models and large model-powered intelligent software,as well as measuring model interpretability and developing more explainable testing approaches,aiming to provide a systematic reference and inspira-tion for subsequent research in this field.