1 迁移学习
表1 传统学习与迁移学习的对比 |
| 学习方式 | Ds&DT | Ts&TT | |
|---|---|---|---|
| 传统学习 | 相同 | 相同 | |
| 归纳迁移学习 | 相同/相关 | 相关 | |
| 迁移学习 | 无监督迁移学习 | 相关 | 相关 |
| 直推式迁移学习 | 相关 | 相同 | |
|
李永盛(1992—),男,河南上蔡县人,硕士研究生,研究方向为迁移学习,视频目标检测等。 |
|
何佳洲(1966—),男,博士生导师,研究员。 |
Copy editor: 胡前进
收稿日期: 2019-10-16
修回日期: 2019-12-03
网络出版日期: 2022-05-10
The Research on Negative Transfer in Transfer Learning
Received date: 2019-10-16
Revised date: 2019-12-03
Online published: 2022-05-10
李永盛 , 何佳洲 , 赵国清 , 刘义海 . 关于迁移学习中的负迁移方向研究[J]. 指挥控制与仿真, 2020 , 42(4) : 28 -33 . DOI: 10.3969/j.issn.1673-3819.2020.04.006
When faced with small sample data, transfer learning has a good processing effect, which is one of the hot research directions of artificial intelligence.However, it is found that there are many problems in the practical application of transfer learning, especially the negative transfer phenomenon, which makes the transfer learning effect not good. In this article, we firstly review the concept of transfer learning and the research status, and then put forward three directions to solve the problem of negative transfer, including multi-source domain data learning, increasing the number of samples in the target domain, and reducing the data differences between domains. And we summarize the related research work to improve negative transfer based on the three directions.Finally, some other directions on improving negative transfer learning are looked forward to, which can provide some references for future generations to further study negative transfer.
Key words: machine learning; transfer learning; negative-transfer
表1 传统学习与迁移学习的对比 |
| 学习方式 | Ds&DT | Ts&TT | |
|---|---|---|---|
| 传统学习 | 相同 | 相同 | |
| 归纳迁移学习 | 相同/相关 | 相关 | |
| 迁移学习 | 无监督迁移学习 | 相关 | 相关 |
| 直推式迁移学习 | 相关 | 相同 | |
| [1] |
|
| [2] |
|
| [3] |
|
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
|
| [8] |
宋鹏, 郑文明, 赵力. 基于特征迁移学习方法的跨库语音情感识别[J]. 清华大学学报(自然科学版), 2016, 56(11):1179-1183.
|
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
|
| [16] |
|
| [17] |
|
| [18] |
季鼎承, 蒋亦樟, 王士同. 基于域与样例平衡的多源迁移学习方法[J]. 电子学报, 2019, 47(3):692-699.
|
| [19] |
刘振, 杨俊安, 刘辉, 等. 基于域相关性与流形约束的多源域迁移学习分类算法[J]. 计算机应用研究, 2017, 34(2):351-356.
|
| [20] |
|
| [21] |
|
| [22] |
|
| [23] |
|
| [24] |
|
| [25] |
|
| [26] |
臧文华. 基于生成对抗网络的迁移学习算法研究[D]. 成都:电子科技大学, 2018.
|
| [27] |
满子健. 基于迁移学习的生成对抗网络算法改进与应用[D]. 徐州:中国矿业大学, 2019.
|
| [28] |
|
| [29] |
|
| [30] |
|
| [31] |
|
| [32] |
|
| [33] |
崔鹏, 赵莎莎. 基于稀疏编码和背景差分的迁移学习行人检测算法[J]. 光电子·激光, 2018, 29(9):1012-1020.
|
| [34] |
|
| [35] |
|
| [36] |
|
| [37] |
Cai, Deng, et al. Graph Regularized Nonnegative Matrix Factorization for Data Representation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 33(8): 1548-1560.
|
| [38] |
龙明盛. 迁移学习问题与方法研究[D]. 北京:清华大学, 2014.
|
| [39] |
|
| [40] |
|
| [41] |
Kim, Tae-Kyun, Roberto, et al. Making a Shallow Network Deep: Conversion of a Boosting Classifier Into a Decision Tree by Boolean Optimisation[J]. International Journal of Computer Vision, 2012, 100(2): 203-215.
|
| [42] |
|
| [43] |
罗娟. 多线索植物种类识别方法研究[D]. 陕西杨凌:西北农林科技大学, 2019.
|
| [44] |
|
| [45] |
|
/
| 〈 |
|
〉 |