| LDR | | 00000nam u2200205 4500 |
| 001 | | 000000434594 |
| 005 | | 20200226160217 |
| 008 | | 200131s2019 ||||||||||||||||| ||eng d |
| 020 | |
▼a 9781687967121 |
| 035 | |
▼a (MiAaPQ)AAI22620704 |
| 040 | |
▼a MiAaPQ
▼c MiAaPQ
▼d 247004 |
| 082 | 0 |
▼a 621 |
| 100 | 1 |
▼a Wu, Yue. |
| 245 | 10 |
▼a Face Recognition by Deep Learning. |
| 260 | |
▼a [S.l.]:
▼b Northeastern University.,
▼c 2019. |
| 260 | 1 |
▼a Ann Arbor:
▼b ProQuest Dissertations & Theses,
▼c 2019. |
| 300 | |
▼a 124 p. |
| 500 | |
▼a Source: Dissertations Abstracts International, Volume: 81-04, Section: B. |
| 500 | |
▼a Advisor: Fu, Yun. |
| 502 | 1 |
▼a Thesis (Ph.D.)--Northeastern University, 2019. |
| 506 | |
▼a This item must not be sold to any third party vendors. |
| 520 | |
▼a This dissertation focuses on real world challenges when applying deep learning for face recognition. Face recognition aims to recognize people using their face images, which is a hot topic in computer vision field. Especially tons of face data exist on social media nowadays and yield tremendous real world applications. We explore five applications along with challenges applying deep learning algorithms. First, we aim to recognize the large number of people, which is the bottleneck for training a deep convolutional neural network of which the output is equal to the number of people. An independent softmax model is introduced to split the single classifier into several small classifiers, which decomposes the large scale training procedure into several medium training procedures which can be solved separately. Second, we study the low-shot face recognition problem, in which there is very limited number of training samples for some people to recognize. A hybrid classifier framework is presented with multiple classifiers to decompose a single classifier into multiple classifiers that each works well for a part of data. Third, feature representation learning with unbalance data is studied for the face verification application. A center-invariant loss is employed to regularize the deep representation learning. Forth, we study the kinship classification given deep representations. A latent adaptive subspace learning framework is proposed to model the family-wise constraint and person-wise constraint in the subspace based on deep representations. Fifth, we study the catastrophic forgetting problem in incremental learning system, especially for face applications with continuously adding more people in the recognition system. A bias correction model is presented together with knowledge distilling, which tackles the catastrophic forgetting problem that has bias towards new coming classes. |
| 590 | |
▼a School code: 0160. |
| 650 | 4 |
▼a Computer engineering. |
| 690 | |
▼a 0464 |
| 710 | 20 |
▼a Northeastern University.
▼b Electrical and Computer Engineering. |
| 773 | 0 |
▼t Dissertations Abstracts International
▼g 81-04B. |
| 773 | |
▼t Dissertation Abstract International |
| 790 | |
▼a 0160 |
| 791 | |
▼a Ph.D. |
| 792 | |
▼a 2019 |
| 793 | |
▼a English |
| 856 | 40 |
▼u http://www.riss.kr/pdu/ddodLink.do?id=T15493752
▼n KERIS
▼z 이 자료의 원문은 한국교육학술정보원에서 제공합니다. |
| 980 | |
▼a 202002
▼f 2020 |
| 990 | |
▼a ***1008102 |
| 991 | |
▼a E-BOOK |