Incomplete Multi-view Clustering via Hierarchical Semantic Alignment and Cooperative Completion

TL;DR

Incomplete Multi-view Clustering via Hierarchical Semantic Alignment and Cooperative Completion (HSACC) introduces a dual-level semantic framework to robustly fuse incomplete views and implicitly complete missing views. By combining low-level mutual information-based consistency with high-level adaptive view weighting and distribution alignment, HSACC achieves strong cross-view coherence and discriminative representations. The method jointly optimizes reconstruction, cross-view inference, and clustering, mitigating error propagation typical of two-stage pipelines. Empirical results on five benchmarks show consistent gains over state-of-the-art IMVC methods, with ablations validating the contribution of each loss component and the robustness to hyperparameters.

Abstract

Incomplete multi-view data, where certain views are entirely missing for some samples, poses significant challenges for traditional multi-view clustering methods. Existing deep incomplete multi-view clustering approaches often rely on static fusion strategies or two-stage pipelines, leading to suboptimal fusion results and error propagation issues. To address these limitations, this paper proposes a novel incomplete multi-view clustering framework based on Hierarchical Semantic Alignment and Cooperative Completion (HSACC). HSACC achieves robust cross-view fusion through a dual-level semantic space design. In the low-level semantic space, consistency alignment is ensured by maximizing mutual information across views. In the high-level semantic space, adaptive view weights are dynamically assigned based on the distributional affinity between individual views and an initial fused representation, followed by weighted fusion to generate a unified global representation. Additionally, HSACC implicitly recovers missing views by projecting aligned latent representations into high-dimensional semantic spaces and jointly optimizes reconstruction and clustering objectives, enabling cooperative learning of completion and clustering. Experimental results demonstrate that HSACC significantly outperforms state-of-the-art methods on five benchmark datasets. Ablation studies validate the effectiveness of the hierarchical alignment and dynamic weighting mechanisms, while parameter analysis confirms the model's robustness to hyperparameter variations.

Figures (5)

• Figure 1: The framework consists of three modules: (a) view reconstruction, where autoencoders extract latent representations and reconstruct the inputs; (b) multi-view representation learning, which aligns views by maximizing mutual information, learns adaptive weighting to fuse complementary information, and minimizes distribution discrepancies; (c) data recovery and clustering, where missing views are completed and clustering is performed based on the completed representations.
• Figure 2: Parameter sensitivity analysis on Caltech101-20 dataset with missing rate 0.5.
• Figure 3: t-SNE visualizations of clustering results on the Caltech101-20 and Noisy MNIST datasets with increasing training iteration. (a)-(d) show the results on the Caltech101-20 dataset. (e)-(h) show the results on the Noisy MNIST dataset.
• Figure 4: Clustering metrics (ACC, NMI, ARI) and loss curves over training epochs on the Caltech101-20, Noisy MNIST, and LandUse_21 datasets.
• Figure 5: t-SNE visualizations of clustering results on the LandUse_21 and Hdigit and 100leaves datasets with increasing training iteration. (a)-(d) show the results on the LandUse_21 dataset. (e)-(h) show the results on the Hdigit dataset. (i)-(j) show the results on the 100leaves dataset.