Building Trusted Artificial Intelligence with Cross-view: Cases Study

Abstract

With the widespread application of artificial intelligence, the safety of artificial intelligence has also attracted people’s attention. In this paper, we propose to construct cross-views on three levels including parameter diversification, sample diversification and algorithm diversification which is able to improve the credibility of artificial intelligence. This paper discusses the difficulties and feasible solutions of the three methods, and illustrates the specific implementation of the three diversification with three cases. In our opinion, artificial intelligence security problems, in a short time, can not be completely solved. Taking diverse approaches and constructing cross-views may be a feasible way to mitigate AI security issues.