Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
SAFETY VERIFICATION SYSTEM FOR ARTIFICIAL INTELLIGENCE SYSTEM, SAFETY VERIFICATION METHOD, AND SAFETY VERIFICATION PROGRAM
Document Type and Number:
WIPO Patent Application WO/2021/100597
Kind Code:
A1
Abstract:
Provided is a system that is effective for verifying the safety of an artificial intelligence system. A safety verification system comprising: a feature amount information reception unit 14 that receives, for each of a plurality of first test data items used in a test for verifying the safety of an artificial intelligence system, feature amount information including the values of a plurality of feature amounts predicted to be used in the artificial intelligence system; and an assessment unit 16 that assesses a first combination, among combinations of values that could be obtained for the plurality of feature amounts, that is not included in the plurality of first test data items, or a second combination, among the combinations of values that could be obtained for the plurality of feature amounts, that is a combination in which a plurality of correct analysis results to be derived by the artificial intelligence are associated.

Inventors:
NOMOTO HIDEKI (JP)
MICHIURA YASUTAKA (JP)
IINO SHOTA (JP)
Application Number:
PCT/JP2020/042229
Publication Date:
May 27, 2021
Filing Date:
November 12, 2020
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
JAPAN MANNED SPACE SYSTEMS CORP (JP)
International Classes:
G06F11/36; G06N20/00
Foreign References:
JP2019075035A2019-05-16
Other References:
SHUBHAM SHARMA; JETTE HENDERSON; JOYDEEP GHOSH: "CERTIFAI: Counterfactual Explanations for Robustness, Transparency, Interpretability, and Fairness of Artificial Intelligence models", ARXIV.ORG [ ONLINE, May 2019 (2019-05-01), XP081662746, Retrieved from the Internet [retrieved on 20200119]
KATAHIRA,MASAFUMI: "Validation technology for upstream process specifications in space systems", SEC JOURNAL,, vol. 10, no. 2, 2 October 2014 (2014-10-02), pages 46 - 53, XP009533718, ISSN: 1349-8622
NOMOTO, HIDEKI; MICHIURA, YASUTAKA; ISHIHAMA, NAOKI; KATAHIRA, MASAFUMI: "New safety engineering using FRAM (functional resonance analysis method)", SEC JOURNAL, vol. 14, no. 1, August 2018 (2018-08-01), Japan, pages 42 - 49, XP009533717, ISSN: 1349-8622
See also references of EP 4002127A4
Attorney, Agent or Firm:
YAMAMOTO, Osamu et al. (JP)
Download PDF: