许聪颖
内容简介
题目
Advancements in Metamorphic Testing: Automated Deduction and Synthesis of Metamorphic Relations
摘要
Metamorphic Testing (MT) has proven invaluable in overcoming the oracle and test case generation problems. Instead of checking individual concrete inputs, MT verifies the behavior of a subject under test against a Metamorphic Relation (MR) that governs multiple related inputs and their outputs. An MR can be applied to a wide range of test inputs, exercising diverse program behaviors without the need to prepare oracles for individual inputs. MT has been successful in detecting critical faults across various software domains, including compilers, databases, and AI-enabled systems. However, constructing MRs is challenging due to the requirement of domain-specific knowledge and reliance on the expertise of testers.
In this talk, I will present a novel approach to synthesize MRs from existing test cases, even if not originally designed for MT. We found that over 11,000 MRs can be discovered and synthesized for new test generation, while over 70% of them lack explicit input relations. To address this gap, we further designed an LLM-based approach to deduce input transformations to complement those MRs. By integrating these MRs with automatically generated inputs, automated MT can be achieved, thereby improving software testing efficiency and adequacy.
报告人
许聪颖,香港科技大学博士生,导师:张成志 (IEEE fellow);于2022年从复旦大学取得硕士学位,导师:陈碧欢、彭鑫;于2019年从扬州大学取得本科学位。目前主要研究方向为软件测试,包括蜕变测试、AI系统测试等研究课题。研究成果发表在ICSE、FSE、TOSEM等软件工程顶级会议和期刊, 并荣获IEEE TCSE杰出论文奖。
时间安排
❖
时间:2024年7月1日,10:00 – 12:00
❖
地点:复旦大学江湾校区交叉二号学科楼A4009