Distilling Influences to Mitigate Prediction Churn in Graph Neural Networks

被引:0
|
作者
Roth, Andreas [1 ]
Liebig, Thomas [1 ,2 ]
机构
[1] TU Dortmund Univ, Dortmund, Germany
[2] Lamarr Inst Machine Learning & Artificial Intelli, Dortmund, Germany
关键词
Graph Neural Networks; Knowledge Distillation; Churn; Instability;
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Models with similar performances exhibit significant disagreement in the predictions of individual samples, referred to as prediction churn. Our work explores this phenomenon in graph neural networks by investigating differences between models differing only in their initializations in their utilized features for predictions. We propose a novel metric called Influence Difference (ID) to quantify the variation in reasons used by nodes across models by comparing their influence distribution. Additionally, we consider the differences between nodes with a stable and an unstable prediction, positing that both equally utilize different reasons and thus provide a meaningful gradient signal to closely match two models even when the predictions for nodes are similar. Based on our analysis, we propose to minimize this ID in Knowledge Distillation, a domain where a new model should closely match an established one. As an efficient approximation, we introduce DropDistillation (DD) that matches the output for a graph perturbed by edge deletions. Our empirical evaluation of six benchmark datasets for node classification validates the differences in utilized features. DD outperforms previous methods regarding prediction stability and overall performance in all considered Knowledge Distillation experiments.
引用
收藏
页数:16
相关论文
共 50 条
  • [1] Distilling Holistic Knowledge with Graph Neural Networks
    Zhou, Sheng
    Wang, Yucheng
    Chen, Defang
    Chen, Jiawei
    Wang, Xin
    Wang, Can
    Bu, Jiajun
    [J]. 2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021), 2021, : 10367 - 10376
  • [2] User Churn Prediction Hierarchical Model Based on Graph Attention Convolutional Neural Networks
    Miao, Mei
    Miao, Tang
    Long, Zhou
    [J]. CHINA COMMUNICATIONS, 2024, 21 (07) : 169 - 185
  • [3] Prediction of Churning Game Users Based on Social Activity and Churn Graph Neural Networks
    Han, You-Jung
    Moon, Jihoon
    Woo, Jiyoung
    [J]. IEEE ACCESS, 2024, 12 : 101971 - 101984
  • [4] User Churn Prediction Hierarchical Model Based on Graph Attention Convolutional Neural Networks
    Mei Miao
    Tang Miao
    Zhou Long
    [J]. ChinaCommunications, 2024, 21 (07) : 169 - 185
  • [5] Customer churn prediction by hybrid neural networks
    Tsai, Chih-Fong
    Lu, Yu-Hsin
    [J]. EXPERT SYSTEMS WITH APPLICATIONS, 2009, 36 (10) : 12547 - 12553
  • [6] Hybrid Artificial Neural Networks Using Customer Churn Prediction
    P. Ramesh
    J. Jeba Emilyn
    V. Vijayakumar
    [J]. Wireless Personal Communications, 2022, 124 : 1695 - 1709
  • [7] Deep Convolutional Neural Networks for Customer Churn Prediction Analysis
    Chouiekh, Alae
    Ibn El Haj, El Hassane
    [J]. INTERNATIONAL JOURNAL OF COGNITIVE INFORMATICS AND NATURAL INTELLIGENCE, 2020, 14 (01) : 1 - 16
  • [8] Hybrid Artificial Neural Networks Using Customer Churn Prediction
    Ramesh, P.
    Jeba Emilyn, J.
    Vijayakumar, V.
    [J]. WIRELESS PERSONAL COMMUNICATIONS, 2022, 124 (02) : 1695 - 1709
  • [9] Adaptive Optimization-Enabled Neural Networks to Handle the Imbalance Churn Data in Churn Prediction
    Garimella, Bharathi
    Prasad, G. V. S. N. R. V.
    Prasad, M. H. M. Krishna
    [J]. INTERNATIONAL JOURNAL OF COMPUTATIONAL INTELLIGENCE AND APPLICATIONS, 2021, 20 (04)
  • [10] On the Prediction Instability of Graph Neural Networks
    Klabunde, Max
    Lemmerich, Florian
    [J]. MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES, ECML PKDD 2022, PT III, 2023, 13715 : 187 - 202