Details

Autor(en) / Beteiligte

• Lin, Jiayu
• Huang, Yixue
• Cai, Lingxi
• Peng, Zihao
• Yang, Zan
• Nai, Wei

Titel

Metamorphic Relationship Generation Based on t-Distribution Zeroth-Order Optimization Method

Ist Teil von

• 2021 14th International Symposium on Computational Intelligence and Design (ISCID), 2021, p.316-320

Ort / Verlag

IEEE

Erscheinungsjahr

2021

Quelle

IEEE Xplore

Beschreibungen/Notizen

• Metamophic testing is an important method in software testing, and the generation of metamorphic relationship is just the key to metamorphic testing. However, there are some difficulties in generating the metamorphic relationship itself. Its loss function has item(s) defined by L1 norm, which means that it contains the absolute value item(s) and makes the whole problem be a non differentiable non-smooth optimization one. Thus, the conventional gradient related line search method cannot be used to find the optimal solution of the metamorphic relationship generation. Therefore, in this paper, a metamorphic relationship generation method based on t-distribution zero-order (ZO, a derivative free optimization algorithm) optimization method has been proposed, the traditional gradient solution process has been replaced by a gradient estimator obtained by the very t-distribution ZO optimization method, and the line search framework is used to solve the optimal solution of the non-smooth optimization loss function in metamorphic relationship generation.