Supervised Fractional-Order Embedding Geometrical Multi-View CCA (SFGMCCA) for Multiple Feature Integration

Keisuke Maeda, Yoshiki Ito, Takahiro Ogawa, Miki Haseyama

Abstract

Techniques for integrating different types of multiple features effectively have been actively studied in recent years. Multiset canonical correlation analysis (MCCA), which maximizes the sum of pairwise correlations of inter-view (i.e., between different features), is one of the powerful methods for integrating different types of multiple features, and various MCCA-based methods have been proposed. This work focuses on a supervised MCCA variant in order to construct a novel effective feature integration framework. In this paper, we newly propose supervised fractional-order embedding geometrical multi-view CCA (SFGMCCA). This method constructs not only the correlation structure but also two types of geometrical structures of intra-view (i.e., within each feature) and inter-view simultaneously, thereby realizing more precise feature integration. This method also supports the integration of small sample and high-dimensional data by using the fractional-order technique. We conducted experiments using four types of image datasets, i.e., MNIST, COIL-20, ETH-80 and CIFAR-10. Furthermore, we also performed an fMRI dataset containing brain signals to verify the robustness. As a result, it was confirmed that accuracy improvements using SFGMCCA were statistically significant at the significance level of 0.05 compared to those using conventional representative MCCA-based methods.

Introduction

In modern data science, particularly within computer vision and medical imaging, researchers frequently encounter "multi-view" data—situations where a single object is described by multiple distinct feature sets. For instance, an image might be characterized by its color distribution, texture patterns, and geometric shapes. Integrating these disparate views into a unified, low-dimensional representation is essential for tasks like classification and retrieval.

Canonical Correlation Analysis (CCA) and its multiset extension (MCCA) have long been the standard for this integration. However, traditional methods struggle with three primary challenges: the "curse of dimensionality" in high-dimensional/small-sample scenarios, the neglect of local manifold structures (local geometry), and the underutilization of class labels in supervised settings. This paper introduces Supervised Fractional-Order Embedding Geometrical Multi-View CCA (SFGMCCA) to address these gaps.

The SFGMCCA Framework

SFGMCCA integrates three innovative components to create a more robust and discriminative feature space:

  1. Supervised Integration via Label Views: Unlike unsupervised CCA, SFGMCCA treats class labels as an additional "view" using one-hot encoding. This forces the resulting embedding to prioritize features that are most relevant for distinguishing between classes.
  2. Fractional-Order Regularization: To combat the curse of dimensionality, the authors employ a fractional-order technique. By using fractional powers ($\xi$ and $\eta$) of the eigenvalues during covariance matrix reconstruction, the model effectively "smooths" the data. This suppresses noise and stabilizes the model when the number of features exceeds the number of samples.
  3. Dual Geometrical Structures: While some methods consider local relationships within a single feature type (intra-view), SFGMCCA simultaneously models relationships across different feature types (inter-view). By constructing graphs for both, the method ensures that similar samples remain close in the integrated space, regardless of the original feature source.

The optimization problem is formulated as a generalized eigenvalue problem, where global correlation is balanced against local geometric constraints using trade-off parameters ($\alpha$ and $\delta$).

Experimental Evaluation

The efficacy of SFGMCCA was validated across five diverse datasets:

  • MNIST & CIFAR-10: Standard benchmarks for handwritten digit and general object recognition.
  • COIL-20 & ETH-80: Datasets focused on multi-angle object views and categorical classification.
  • fMRI Dataset: A complex dataset pairing brain activity signals with visual features from ImageNet, testing the model's robustness beyond standard image data.

SFGMCCA was compared against eight representative methods, including standard MCCA, Robust MCCA (RMCCA), and Graph-regularized MCCs (GrMCCs).

Key Results and Discussion

The experimental results demonstrate that SFGMCCA consistently outperforms state-of-the-art methods. Notably, on the CIFAR-10 dataset, SFGMCCA achieved an accuracy of 48.2%, a significant improvement over the next best method, SFEMCCA, which reached 42.1%.

Furthermore, the method exhibits high efficiency in low-dimensional spaces, often reaching peak performance with as few as 7 to 9 dimensions. Statistical analysis using Welch’s t-test confirmed that these improvements are significant at the $p < 0.05$ level. The "fractional-order" approach proved particularly effective as a regularizer, preventing the model from over-fitting to the largest, often noise-contaminated, principal components.

Conclusion

SFGMCCA provides a powerful framework for multi-view feature integration, particularly for high-dimensional data with limited samples. By combining supervised learning, fractional-order smoothing, and dual-geometry preservation, it captures a more holistic representation of data than traditional MCCA variants. Future work will explore its application in cross-modal retrieval and further refine its geometric modeling to handle increasingly noisy real-world environments.

Summary updated

Abstract

Techniques for integrating different types of multiple features effectively have been actively studied in recent years. Multiset canonical correlation analysis (MCCA), which maximizes the sum of pairwise correlations of inter-view (i.e., between different features), is one of the powerful methods for integrating different types of multiple features, and various MCCA-based methods have been proposed. This work focuses on a supervised MCCA variant in order to construct a novel effective feature integration framework. In this paper, we newly propose supervised fractional-order embedding geometrical multi-view CCA (SFGMCCA). This method constructs not only the correlation structure but also two types of geometrical structures of intra-view (i.e., within each feature) and inter-view simultaneously, thereby realizing more precise feature integration. This method also supports the integration of small sample and high-dimensional data by using the fractional-order technique. We conducted experiments using four types of image datasets, i.e., MNIST, COIL-20, ETH-80 and CIFAR-10. Furthermore, we also performed an fMRI dataset containing brain signals to verify the robustness. As a result, it was confirmed that accuracy improvements using SFGMCCA were statistically significant at the significance level of 0.05 compared to those using conventional representative MCCA-based methods.
BibTeX
@article{maeda2020supervised,
  author = {Keisuke Maeda and Yoshiki Ito and Takahiro Ogawa and Miki Haseyama},
  title = {{Supervised Fractional-Order Embedding Geometrical Multi-View CCA (SFGMCCA) for Multiple Feature Integration}},
  journal = {IEEE Access},
  year = {2020},
  doi = {10.1109/access.2020.3003619},
}
Figure