Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Regularized Contrastive Partial Multi-view Outlier Detection

Yijia Wang, Qianqian Xu, Yangbangyan Jiang, Siran Dai, Qingming Huang

TL;DR

This work proposes a novel method named Regularized Contrastive Partial Multi-view Outlier Detection (RCPMOD), which utilizes contrastive learning to learn view-consistent information and distinguish outliers by the degree of consistency.

Abstract

In recent years, multi-view outlier detection (MVOD) methods have advanced significantly, aiming to identify outliers within multi-view datasets. A key point is to better detect class outliers and class-attribute outliers, which only exist in multi-view data. However, existing methods either is not able to reduce the impact of outliers when learning view-consistent information, or struggle in cases with varying neighborhood structures. Moreover, most of them do not apply to partial multi-view data in real-world scenarios. To overcome these drawbacks, we propose a novel method named Regularized Contrastive Partial Multi-view Outlier Detection (RCPMOD). In this framework, we utilize contrastive learning to learn view-consistent information and distinguish outliers by the degree of consistency. Specifically, we propose (1) An outlier-aware contrastive loss with a potential outlier memory bank to eliminate their bias motivated by a theoretical analysis. (2) A neighbor alignment contrastive loss to capture the view-shared local structural correlation. (3) A spreading regularization loss to prevent the model from overfitting over outliers. With the Cross-view Relation Transfer technique, we could easily impute the missing view samples based on the features of neighbors. Experimental results on four benchmark datasets demonstrate that our proposed approach could outperform state-of-the-art competitors under different settings.

Regularized Contrastive Partial Multi-view Outlier Detection

TL;DR

This work proposes a novel method named Regularized Contrastive Partial Multi-view Outlier Detection (RCPMOD), which utilizes contrastive learning to learn view-consistent information and distinguish outliers by the degree of consistency.

Abstract

In recent years, multi-view outlier detection (MVOD) methods have advanced significantly, aiming to identify outliers within multi-view datasets. A key point is to better detect class outliers and class-attribute outliers, which only exist in multi-view data. However, existing methods either is not able to reduce the impact of outliers when learning view-consistent information, or struggle in cases with varying neighborhood structures. Moreover, most of them do not apply to partial multi-view data in real-world scenarios. To overcome these drawbacks, we propose a novel method named Regularized Contrastive Partial Multi-view Outlier Detection (RCPMOD). In this framework, we utilize contrastive learning to learn view-consistent information and distinguish outliers by the degree of consistency. Specifically, we propose (1) An outlier-aware contrastive loss with a potential outlier memory bank to eliminate their bias motivated by a theoretical analysis. (2) A neighbor alignment contrastive loss to capture the view-shared local structural correlation. (3) A spreading regularization loss to prevent the model from overfitting over outliers. With the Cross-view Relation Transfer technique, we could easily impute the missing view samples based on the features of neighbors. Experimental results on four benchmark datasets demonstrate that our proposed approach could outperform state-of-the-art competitors under different settings.
Paper Structure (16 sections, 1 theorem, 12 equations, 4 figures, 5 tables)

This paper contains 16 sections, 1 theorem, 12 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-view Outlier Detection
  4. Contrastive Multi-view Learning
  5. Methodology
  6. Problem Setting
  7. Outlier-aware Contrastive Learning
  8. Neighbor Alignment Contrastive Learning
  9. Spreading Regularization
  10. Outlier Scoring
  11. Experiments
  12. Experimental Settings
  13. Comparisons with Baseline Methods
  14. Sensitivity Analysis
  15. Ablation Study
  16. ...and 1 more sections

Key Result

Proposition 1

If $I(\bm{x}_o^{(1)}, \bm{x}_o^{(2)}) \leq \varepsilon$, then the contrastive loss value of outlier instances is lower-bounded by $\log (2N) - \varepsilon$.

Figures (4)

  • Figure 1: Different types of outliers in complete and partial multi-view data. Dashed circles represent the missing views.
  • Figure 2: Overview of RCPMOD on bi-view data. Two key contrastive learning modules are applied on the latent space to promote the view consistency: (1) In outlier-aware contrastive module, potential class-related outliers are restored in a memory bank and used as additional negative samples. (2) In neighbor alignment contrastive module, the corresponding neighbors of a sample are aligned to learn the cross-view structural correlations. Moreover, we adopt a spreading regularization to prevent from overfitting on class-related outliers. The missing samples are imputed by the Cross-view Relation Transfer technique.
  • Figure 3: (a) Comparison of the detection AUC with and without spreading regularization (SR) on SCENE15. (b) Comparison of the average loss value over inliers and outliers. (c)/(d) Outlier score distribution without/with SR.
  • Figure 4: Sensitivity analysis over $\lambda_1$, $\lambda_2$$\eta$ and $\mu$ on different datasets.

Theorems & Definitions (1)

  • Proposition 1