Multi-view Feature Extraction based on Triple Contrastive Heads
This is an incremental improvement for multi-view feature extraction in machine learning.
The paper tackles the problem of high-dimensional multi-view data by proposing a multi-view feature extraction method using triple contrastive heads, which combines sample-, recovery-, and feature-level contrastive losses to extract sufficient yet minimal subspace discriminative information, and numerical experiments show it offers a strong advantage.
Multi-view feature extraction is an efficient approach for alleviating the issue of dimensionality in highdimensional multi-view data. Contrastive learning (CL), which is a popular self-supervised learning method, has recently attracted considerable attention. In this study, we propose a novel multi-view feature extraction method based on triple contrastive heads, which combines the sample-, recovery- , and feature-level contrastive losses to extract the sufficient yet minimal subspace discriminative information in compliance with information bottleneck principle. In MFETCH, we construct the feature-level contrastive loss, which removes the redundent information in the consistency information to achieve the minimality of the subspace discriminative information. Moreover, the recovery-level contrastive loss is also constructed in MFETCH, which captures the view-specific discriminative information to achieve the sufficiency of the subspace discriminative information.The numerical experiments demonstrate that the proposed method offers a strong advantage for multi-view feature extraction.