摘要: 本贴列出实验室在半监督回归方向的一些工作. 主要目的是保存 bib 方便以后查阅与引用.
1. SCI 期刊论文
DTA, Accepted, to appear in 2023.
2. 国际会议论文
2.1 自步安全协同回归
Self-paced safe co-training for regression (SPOR) 算法将自步协同算法与安全学习的理念用于半监督回归领域中. 在无标签数据学习时, SPOR 算法使用自步协同的学习范式, 按照由易到难的方式为协同协同模型挑选最合适的无标签数据. 在挑选的无标签数据之上, 使用安全学习为无标签数据赋予一个性能不会弱于传统伪标签的安全标签用于模型训练.
@inproceedings{MinLiLiu2022Spor,
author = {Fan Min and Yu Li and Liyan Liu},
title = {Self-paced safe co-training for regression},
booktitle = {PAKDD},
year = {2022},
pages = {71--82}
doi = {10.1007/978-3-031-05936-0_6}
}
2.2 安全多视图协同回归
旨在提升伪标签质量的协同训练算法. 该算法基于一致性假设挑选置信样本, 之后利用安全学习为其赋予可靠的伪标签.
@inproceedings{LiuHuangMin2022Safe,
author = {Liyan Liu and Peng Huang and Fan Min},
title = {Safe multi-view co-training for semi-supervised regression},
booktitle = {DSAA},
year = {2022},
pages = {xxx},
doi = {xxx}
}
2.3 数据分区和特征映射回归
考虑将回归器作为分区的标准对数据进行分区回归, 利用最近邻的思想判断未标记数据所属的分区.
@inproceedings{LiuZhangMin2022Partition,
author = {Liyan Liu and Jiahui Zhang and Fan Min},
title = {Semi-supervised regression with data partitioning and feature mapping},
booktitle = {DSAA},
year = {2022},
pages = {xxx},
doi = {xxx}
}