Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond Correlation: Causal Multi-View Unsupervised Feature Selection Learning

Zongxin Shen, Yanyong Huang, Bin Wang, Jinyuan Chang, Shiyu Liu, Tianrui Li

TL;DR

This work tackles MUFS in unlabeled multi-view data by questioning the reliability of correlations used for feature selection and proposing a causal framework. It introduces a structural causal model to reveal how confounders can induce spurious associations and then presents CAUSA, which couples a generalized unsupervised spectral regression with a causal regularization module that adaptively separates confounders and learns view-shared sample weights. The method jointly optimizes feature selection and confounder balancing, enabling identification of causally informative features across views; an efficient alternating optimization algorithm is developed. Extensive experiments on six real multi-view datasets and synthetic data show that CAUSA outperforms state-of-the-art MUFS methods and demonstrates the value of enforcing causal reasoning in unsupervised, multi-view feature selection.

Abstract

Multi-view unsupervised feature selection (MUFS) has recently received increasing attention for its promising ability in dimensionality reduction on multi-view unlabeled data. Existing MUFS methods typically select discriminative features by capturing correlations between features and clustering labels. However, an important yet underexplored question remains: \textit{Are such correlations sufficiently reliable to guide feature selection?} In this paper, we analyze MUFS from a causal perspective by introducing a novel structural causal model, which reveals that existing methods may select irrelevant features because they overlook spurious correlations caused by confounders. Building on this causal perspective, we propose a novel MUFS method called CAusal multi-view Unsupervised feature Selection leArning (CAUSA). Specifically, we first employ a generalized unsupervised spectral regression model that identifies informative features by capturing dependencies between features and consensus clustering labels. We then introduce a causal regularization module that can adaptively separate confounders from multi-view data and simultaneously learn view-shared sample weights to balance confounder distributions, thereby mitigating spurious correlations. Thereafter, integrating both into a unified learning framework enables CAUSA to select causally informative features. Comprehensive experiments demonstrate that CAUSA outperforms several state-of-the-art methods. To our knowledge, this is the first in-depth study of causal multi-view feature selection in the unsupervised setting.

Beyond Correlation: Causal Multi-View Unsupervised Feature Selection Learning

TL;DR

This work tackles MUFS in unlabeled multi-view data by questioning the reliability of correlations used for feature selection and proposing a causal framework. It introduces a structural causal model to reveal how confounders can induce spurious associations and then presents CAUSA, which couples a generalized unsupervised spectral regression with a causal regularization module that adaptively separates confounders and learns view-shared sample weights. The method jointly optimizes feature selection and confounder balancing, enabling identification of causally informative features across views; an efficient alternating optimization algorithm is developed. Extensive experiments on six real multi-view datasets and synthetic data show that CAUSA outperforms state-of-the-art MUFS methods and demonstrates the value of enforcing causal reasoning in unsupervised, multi-view feature selection.

Abstract

Multi-view unsupervised feature selection (MUFS) has recently received increasing attention for its promising ability in dimensionality reduction on multi-view unlabeled data. Existing MUFS methods typically select discriminative features by capturing correlations between features and clustering labels. However, an important yet underexplored question remains: \textit{Are such correlations sufficiently reliable to guide feature selection?} In this paper, we analyze MUFS from a causal perspective by introducing a novel structural causal model, which reveals that existing methods may select irrelevant features because they overlook spurious correlations caused by confounders. Building on this causal perspective, we propose a novel MUFS method called CAusal multi-view Unsupervised feature Selection leArning (CAUSA). Specifically, we first employ a generalized unsupervised spectral regression model that identifies informative features by capturing dependencies between features and consensus clustering labels. We then introduce a causal regularization module that can adaptively separate confounders from multi-view data and simultaneously learn view-shared sample weights to balance confounder distributions, thereby mitigating spurious correlations. Thereafter, integrating both into a unified learning framework enables CAUSA to select causally informative features. Comprehensive experiments demonstrate that CAUSA outperforms several state-of-the-art methods. To our knowledge, this is the first in-depth study of causal multi-view feature selection in the unsupervised setting.

Paper Structure

This paper contains 12 sections, 1 theorem, 11 equations, 8 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Causal Analysis on MUFS
  3. Notations and Problem Definition
  4. Causal Perspective on MUFS
  5. The Proposed Method
  6. Formulation of CAUSA
  7. Optimization and Analyses
  8. Experiments
  9. Experimental Settings
  10. Experiments on Synthetic Dataset
  11. Experiments on Real Multi-view Datasets
  12. Conclusion

Key Result

Theorem 1

Suppose the confounder $\mathcal{C}^{(v)}$ is properly adjusted so that, under the adjusted distribution $\tilde{P}$, it is independent of the non-causal features $\mathcal{U}^{(v)}$, i.e., $\tilde{P}(\mathcal{C}^{(v)} \mid \mathcal{U}^{(v)} = u^{(v)}) = \tilde{P}(\mathcal{C}^{(v)})$ for all $u^{(v)

Figures (8)

  • Figure 1: The framework of the proposed CAUSA.
  • Figure 2: The SCM graph for MUFS.
  • Figure 3: Feature selection results of CAUSA and the runner-up method on the synthetic dataset. Yellow area: causal feature indices; others: non-causal feature indices.
  • Figure 4: ACC of different methods on six multi-view datasets under different feature selection ratios.
  • Figure 5: NMI of different methods on six multi-view datasets under different feature selection ratios.
  • ...and 3 more figures

Theorems & Definitions (1)

  • Theorem 1