Multi-View Semi-Supervised Label Distribution Learning with Local Structure Complementarity

TL;DR

This work tackles label distribution learning (LDL) in a multi-view semi-supervised setting. It introduces MVSS-LDL, a graph-based model that jointly learns a fused similarity graph, per-view label distributions, and view-specific projections while enforcing cross-view consistency and cross-view neighbor complementarity via a complemented neighbourhood $oldsymbol{ abla}(oldsymbol{x}_i)=\bigcup_v \boldsymbol{ abla}(oldsymbol{x}_i^v)$. The optimization alternates among updating the similarity matrix $f S$, the predicted distribution matrix $f D$, and the projection matrices $f W^v$, with test-time predictions averaged over views. Experiments on six multi-view LDL datasets show MVSS-LDL consistently outperforms single-view LDL baselines under limited labeled data, and ablation confirms the benefit of both the similarity and label distribution consistency terms. The approach offers a principled way to exploit multi-view data and unlabeled samples for more accurate LDL in practical applications.

Abstract

Label distribution learning (LDL) is a paradigm that each sample is associated with a label distribution. At present, the existing approaches are proposed for the single-view LDL problem with labeled data, while the multi-view LDL problem with labeled and unlabeled data has not been considered. In this paper, we put forward the multi-view semi-supervised label distribution learning with local structure complementarity (MVSS-LDL) approach, which exploits the local nearest neighbor structure of each view and emphasizes the complementarity of local nearest neighbor structures in multiple views. Specifically speaking, we first explore the local structure of view $v$ by computing the $k$-nearest neighbors. As a result, the $k$-nearest neighbor set of each sample $\boldsymbol{x}_i$ in view $v$ is attained. Nevertheless, this $k$-nearest neighbor set describes only a part of the nearest neighbor information of sample $\boldsymbol{x}_i$. In order to obtain a more comprehensive description of sample $\boldsymbol{x}_i$'s nearest neighbors, we complement the nearest neighbor set in view $v$ by incorporating sample $\boldsymbol{x}_i$'s nearest neighbors in other views. Lastly, based on the complemented nearest neighbor set in each view, a graph learning-based multi-view semi-supervised LDL model is constructed. By considering the complementarity of local nearest neighbor structures, different views can mutually provide the local structural information to complement each other. To the best of our knowledge, this is the first attempt at multi-view LDL. Numerical studies have demonstrated that MVSS-LDL attains explicitly better classification performance than the existing single-view LDL methods.

Figures (5)

• Figure 1: An example of label distributions. The distribution degrees are 0.665 for “Mountain” (Mou), 0.213 for “Sky”, and 0.122 for “Cloud” (Clo). The other labels are 0.
• Figure 2: Sample images from (a) SCUT-FBP, (b) Emotion6, (c) fbp5500, (d) Twitter-LDL, and (e) Flickr-LDL.
• Figure 3: The Chebyshev value ($\downarrow$) of MVSS-LDL and the baselines with different number of labeled samples.
• Figure 4: The Chebyshev value ($\downarrow$) of MVSS-LDL with different value of parameter $\lambda$.
• Figure 5: The results of ablation experiments under all evaluation metrics.