A Confidence Index Based Test Approach for AI Systems
The tutorial is based on our research and experiments with Buick LaVida effective real claim cases for car damage, we designed an intelligent claim evaluation generation model based on a multi-layer deep learning algorithm. Addition to the traditional risk-based analysis testing, we designed a confidence index approach to manage the risk associated with the “black-box” nature of the AI system. The proposed confidence index could be computed as either statistic or machine learning methods, in which considering the following factors, such as
1) historical data coefficient of variation,
2) data convergence degree, and
3) data item similarity.
It has been applied to predict the automotive body parts replacement and labor cost based on deep learning algorithms, which effectively control the potential risks by unrecognized cases caused AI system failure.
Professor Dr. Qin Liu
Professor Dr. Liu has been a co-founder and the national representative of the Chinese Software Testing Qualifications Board since 2006, and a co-editor of ISO29119 International Software Testing Standard since 2011. She has been serving the school of software at Tongji University as the Executive Dean since 2008 and leading the school to the top 10% ranking in the national evaluation by Ministry of Education in China in 2017. Professor Liu is a leader in machine learning with success creating state-of-the-art solutions that potentially impact real world inventions. Her research focuses on processing and prediction based on heterogeneous data and transferring machine-learning methods to solutions solving multi-disciplinary problems. she has been leading over 10 research projects which have strong links to the industry and local government departments, and had 30 publications in the field of big data oriented, AI based algorithms and models.