지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학위논문
- 저자정보
- 지도교수
- 김명주
- 발행연도
- 2019
- 저작권
- 서울여자대학교 논문은 저작권에 의해 보호받습니다.
이용수74
이 논문의 연구 히스토리 (2)
- 이 논문의 후속연구가 궁금하신가요?
- 연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
- 이 논문의 연구 히스토리 확인하기
초록· 키워드
오류제보하기
With the development of artificial intelligence, artificial intelligence based decision making system is used in various fields of society. Although AI technology is applied to various service, artificial intelligence is based on the black box model so we can not know how or why we reached the result. Because artificial intelligence learns past data, if the data are biased then AI learning result can also be biased. To get reliable AI, it is required these characteristics such as fairness, accountability, transparency. In this paper, we propose a methodology to mitigate biases in using AI. The proposed method is to mitigate biases by distinguishing previous studies that define fairness using statistical approaches and machine learning structured approaches. At first at the pre-processing stage, AI learns from the biased data and statistical treatment data each. And then at the in-processing stage, the threshold of the biased data are compared with the threshold of the mitigated data. The process of treating again and comparing data are repeated until the difference between the biased data and the previous data is minimized. At the final stage of post-processing, the characteristics that are not relevant to decision-making results are removed by counterfactual analysis. The methodology proposed in this study repeated the process of comparing the biased dataset with the de-biased dataset to mitigate bias. Based on this, it is possible to contribute to the development of artificial intelligence technology by mitigating the bias that can occur when using artificial intelligence.
목차
제 1 장 서 론 ······························································································ 11.1. 연구배경 및 필요성 ··················································································· 11.2. 연구 내용 ····································································································· 21.3. 논문의 구성 ································································································· 3제 2 장 관련 연구 ······················································································ 42.1. 인공지능의 활용 ························································································· 42.2. 편향성의 정의 ····························································································· 52.3. 편향성 분석 방법론 ··················································································· 6제 3 장 심층 사례 연구 ·········································································· 113.1. Propublica의 COMPAS 재분석 ································································ 113.2. 기술 현황 분석 ························································································· 113.2.1. IBM AIF360 ·························································································· 123.2.2. Google What-If Tool ········································································· 133.2.3. Microsoft FairLearn ··········································································· 143.2.4. Themis-ML ·························································································· 143.2.5. 기술 현황 분석 결과 ········································································ 15제 4 장 본 론 ···························································································· 174.1. 편향성이 완화된 인공지능 활용을 위한 연구 방향 ························· 174.2. 편향성 완화 방법론 ················································································· 19제 5 장 실험결과 ······················································································ 25제 6 장 결 론 ·························································································· 28참 고 문 헌 ······························································································ 29부 록 ·········································································································· 31ABSTRACT ································································································ 40
최근 본 자료
댓글(0)
0