DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Shin, Jinwoo | - |
dc.contributor.advisor | 신진우 | - |
dc.contributor.author | Kim, Jaehyung | - |
dc.date.accessioned | 2019-09-04T02:42:35Z | - |
dc.date.available | 2019-09-04T02:42:35Z | - |
dc.date.issued | 2019 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=867969&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/266828 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전기및전자공학부, 2019.2,[iii, 18 p. :] | - |
dc.description.abstract | A large variety of labeled datasets in practice are often severely imbalanced class-wise and it is well-known that modern deep neural networks poorly generalize from such datasets due to overfitting classes of majority training samples. To address the issue, we explore the recent state-of-the-art regularization method, called Mixup, and found that it is also very effective for class imbalanced training. Motivated by this, we propose a novel, yet simple advanced alternative, coined Boundary-Mixup. It also generates synthetic training samples by following the original Mixup principle, but our main idea beyond it is to balance the uncertainty level between classes by generating mixed samples near the decision boundary of the classifier. We demonstrate the effectiveness of Boundary-Mixup for image classification, natural language processing and continual learning tasks, improving the prior baseline methods including the original Mixup. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | data imbalance▼aregularization▼adeep learning▼amachine learning | - |
dc.subject | mix-up | - |
dc.subject | 데이터 불균형▼a일반화▼a딥러닝▼a머신러닝▼a믹스업 | - |
dc.title | Improving deep imbalanced classification via large-margin mix-up | - |
dc.title.alternative | 여분 최대화 믹스업을 통한 불균형 심층 분류의 성능 개선 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :전기및전자공학부, | - |
dc.contributor.alternativeauthor | 김재형 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.