| 000 | 00000nam c2200205 c 4500 | |
| 001 | 000046026274 | |
| 005 | 20230526122034 | |
| 007 | ta | |
| 008 | 200107s2020 ulkad bmAC 000c eng | |
| 040 | ▼a 211009 ▼c 211009 ▼d 211009 | |
| 041 | 0 | ▼a eng ▼b kor |
| 085 | 0 | ▼a 0510 ▼2 KDCP |
| 090 | ▼a 0510 ▼b 6D36 ▼c 1113 | |
| 100 | 1 | ▼a 박주연, ▼g 朴珠延 |
| 245 | 1 0 | ▼a REFGAN : ▼b understanding deep network decisions via reference generation / ▼d Jouyon Park |
| 260 | ▼a Seoul : ▼b Graduate School, Korea University, ▼c 2020 | |
| 300 | ▼a iv, 22장 : ▼b 천연색삽화, 도표 ; ▼c 26 cm | |
| 500 | ▼a 지도교수: 이성환 | |
| 502 | 0 | ▼a 학위논문(석사)-- ▼b 고려대학교 대학원, ▼c 컴퓨터·전파통신공학과, ▼d 2020. 2 |
| 504 | ▼a 참고문헌: 장 21-22 | |
| 530 | ▼a PDF 파일로도 이용가능; ▼c Requires PDF file reader(application/pdf) | |
| 653 | ▼a Computer Science | |
| 776 | 0 | ▼t REFGAN: Understanding Deep Network Decisions via Reference Generation ▼w (DCOLL211009)000000127348 |
| 900 | 1 0 | ▼a Park, Jou-yon, ▼e 저 |
| 900 | 1 0 | ▼a 이성환, ▼g 李晟瑍, ▼d 1962-, ▼e 지도교수 ▼0 AUTH(211009)151678 |
| 945 | ▼a KLPA |
전자정보
소장정보
| No. | 소장처 | 청구기호 | 등록번호 | 도서상태 | 반납예정일 | 예약 | 서비스 |
|---|---|---|---|---|---|---|---|
| No. 1 | 소장처 과학도서관/학위논문서고/ | 청구기호 0510 6D36 1113 | 등록번호 123063747 | 도서상태 대출가능 | 반납예정일 | 예약 | 서비스 |
| No. 2 | 소장처 과학도서관/학위논문서고/ | 청구기호 0510 6D36 1113 | 등록번호 123063748 | 도서상태 대출가능 | 반납예정일 | 예약 | 서비스 |
컨텐츠정보
초록
Abstract Due to the complex nature of machine learning models, it has been hard to understand how networks behave, and why they fail. Recent work on explaining deep network decisions provide visual hints which show which feature in the input image deep architectures focused on. However, these visual explanations only give rough semantic explanations such as high-lighting important spots or features. These explanations are often not sufficient for people to understand, and furthermore interact with machine learning models. In this paper, we propose a framework to explain deep network decisions using reference generation based on the training data distribution. We give rich example images which show the direct example data in the training set that influenced the classification process, and may be used to retrain the classifier when the user does not agree with the data point.
목차
1 Introduction 1 2 Related Works 6 3 Materials and Methods 9 3.1 Generative Adversarial Networks (GAN) 10 3.2 Searching Latent Space 12 3.3 Finding Reference Images 13 4 Results and Analysis 15 4.1 Image Analysis 15 4.2 Classifier Analysis 17 4.3 Retraining Scheme 18 5 Discussion and Conclusions 20 6 Reference 21
