基本信息
浏览量:5
职业迁徙
个人简介
Graph neural networks are widely utilized for addressing common real-world problems such as classification, clustering, and link prediction. However, deep graph neural networks often suffer from performance degradation. Previous solutions have mainly focused on mitigating over-smoothing, while our approach specifically targets the issue of over-correlation. We observed that previous methods reduce over-correlation between feature dimensions by influencing the model weights. This led us to investigate whether feature propagation plays a crucial role in feature over-correlation. Our findings confirmed that feature propagation is indeed a key factor in this regard. To address this problem, we propose a novel decorrelation propagation method based on graph signal denoising optimization with orthogonal regularization, which is applied to the classical graph neural network models. Experimental results demonstrate the effectiveness of our method and its strong performance across diverse datasets.
研究兴趣
论文共 4 篇作者统计合作学者相似作者
按年份排序按引用量排序主题筛选期刊级别筛选合作者筛选合作机构筛选
时间
引用量
主题
期刊级别
合作者
合作机构
PROCEEDINGS OF THE 29TH ACM SIGKDD CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, KDD 2023pp.1466-1476, (2023)
Future Generation Computer Systems (2020): 403-408
作者统计
合作学者
合作机构
D-Core
- 合作者
- 学生
- 导师
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn