DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Chung, Hye Won | - |
dc.contributor.advisor | 정혜원 | - |
dc.contributor.author | Kim, Haeri | - |
dc.date.accessioned | 2022-04-27T19:31:38Z | - |
dc.date.available | 2022-04-27T19:31:38Z | - |
dc.date.issued | 2021 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=948699&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/296058 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전기및전자공학부, 2021.2,[iv, 33 p. :] | - |
dc.description.abstract | In the real world, the dataset used to train generative models may include noisy samples. This hinders the model from learning the desired distribution. We propose a method to make robust Generative Adversarial Networks (GANs). We empirically analyze the unstable behavior of GAN training on hard-to-be-learned data and develop efficient measures Log Density Ratio Variance (LDRV) and Log Density Ratio Difference (LDRD) to identify these samples. Using the property that noisy data is hard-to-be-learned, LDRV and LDRD can be used to filter out the noisy data in the training dataset. Furthermore, for clean datasets, forcing the model to focus on hard-to-be-learned data by sample re-weighting using LDRV and LDRD can enhance training performance. Finally, we can make a robust GAN by filtering noisy data and re-weighting. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | Generative Adversarial Network(GAN)▼aRobust training▼aNoise data filtering▼aSample re-weighting▼aLearning dynamics | - |
dc.subject | 생성적 적대 신경망▼a강인한 학습▼a잡음 데이터 제거▼a데이터 가중치 조절▼a학습 역학 | - |
dc.title | Diagnosing data samples for robust GAN training | - |
dc.title.alternative | 강인한 생성적 적대 신경망 학습을 위한 데이터 샘플 진단 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :전기및전자공학부, | - |
dc.contributor.alternativeauthor | 김해리 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.