Research map / Federated learning and privacy / Personalized and heterogeneous FL
Efficient Federated Incomplete Multi-View Clustering
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes EFIMVC, a federated incomplete multi-view clustering framework that optimizes locally and communicates only view-specific anchor graphs and anchor similarity matrices, handling missing views via index matrices.
Problem
Federated multi-view clustering suffers from high communication overhead, privacy leakage from transmitting embeddings or sample-level similarity matrices, and poor handling of missing views.
Approach
Decouples client and server optimization so variables are sent only after local convergence, uses view-specific anchor graphs as communication variables instead of embeddings, and aligns anchor graphs on the server with row-level (alignment matrix) and anchor-similarity terms, with index matrices for missing views.
Key results
- EFIMVC outperforms the best incomplete MVC baseline by 35.6% ACC on 100Leaves and 5.5% on YTF10.
- On ProteinFold and WebKB, EFIMVC beats the second-best federated baseline by 32.9% and 19.7% ACC.
- Objective values for client, server and overall converge within 30 iterations, with the overall objective converging fastest.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use EFIMVC instead of the strongest listed baseline | EFIMVC · ProteinFold | ACC (%): 32.36 -> 31.71 | Hurt |
| Use EFIMVC instead of the strongest listed baseline | EFIMVC · WebKB | ACC (%): 91.51 -> 90.64 | Hurt |
| Use EFIMVC instead of the strongest listed baseline | EFIMVC · 100Leaves | ACC (%): 53.73 -> 72.88 | Helped |
| Use EFIMVC instead of the strongest listed baseline | EFIMVC · CCV | ACC (%): 20.98 -> 20.04 | Hurt |
| Use EFIMVC instead of the strongest listed baseline | EFIMVC · YTF10 | ACC (%): 80.55 -> 84.95 | Helped |
| Use EFIMVC instead of the strongest listed baseline | EFIMVC · CIFAR10 | ACC (%): 96.56 -> 96.62 | Helped |
| Use EFIMVC instead of the strongest listed baseline | EFIMVC · MNIST | ACC (%): 98.70 -> 98.66 | Hurt |
| Add the regularization term (lambda) (ablation) | EFIMVC · ProteinFold, WebKB, 100Leaves, CCV, YTF10, CIFAR10, MNIST | ACC (%): 30.80 -> 31.71; 78.68 -> 90.64; 64.09 -> 72.88; 17.96 -> 20.04; 79.44 -> 84.95; 96.01 -> 96.62; 98.51 -> 98.66 | Helped |
| Add the feature alignment term (beta) (ablation) | EFIMVC · ProteinFold, WebKB, 100Leaves, CCV, YTF10, CIFAR10, MNIST | ACC (%): 30.85 -> 31.71; 90.36 -> 90.64; 68.12 -> 72.88; 14.85 -> 20.04; 78.42 -> 84.95; 96.54 -> 96.62; 98.64 -> 98.66 | Helped |
| Add anchor alignment (learn P(v) rather than fixing P(v)=I) (ablation) | EFIMVC · ProteinFold, WebKB, 100Leaves, CCV, YTF10, CIFAR10, MNIST | ACC (%): 20.18 -> 31.71; 73.61 -> 90.64; 54.46 -> 72.88; 14.82 -> 20.04; 77.04 -> 84.95; 82.70 -> 96.62; 91.25 -> 98.66 | Helped |
| Remove the feature alignment term by setting beta=0 (ablation) | EFIMVC · ProteinFold | ACC (%): 31.71 -> 30.85 | Hurt |
| Remove the feature alignment term by setting beta=0 (ablation) | EFIMVC · CIFAR10, MNIST | ACC (%): 96.62 -> 96.54; 98.66 -> 98.64 | No clear effect |
Limitations
- Missing views handled by zero-filling for federated baselines, making comparisons nonuniform.
- Hyperparameters lambda and beta require tuning per dataset.
- No quantitative communication-cost measurements reported.
Experimental setup
Datasets: ProteinFold WebKB 100Leaves CCV YTF10 CIFAR10 MNIST
Benchmarks: ProteinFold WebKB 100Leaves CCV YTF10 CIFAR10 MNIST
Metrics: ACC NMI Purity Fscore
Baselines: FIMVC SCBGL DSVAI DAQINT FMVC HFMVC FMCSC FedMVFPC
Builds on: anchor-based multi-view clustering FIMVC federated multi-view clustering
Closest papers by meaning
- An Effective and Secure Federated Multi-View Clustering Method with Information-Theoretic Perspective (ICML 2025 · similarity 0.93)
- Federated Incomplete Multi-view Clustering with Globally Fused Graph Guidance (ICML 2025 · similarity 0.91)
- Scalable Federated One-Step Multi-View Clustering with Tensorized Regularization (AAAI 2025 · similarity 0.90)
- Graph Consistency and Diversity Measurement for Federated Multi-View Clustering (AAAI 2025 · similarity 0.90)
- Towards Resource-friendly, Extensible and Stable Incomplete Multi-view Clustering (ICML 2024 · similarity 0.89)
- Bridging Gaps: Federated Multi-View Clustering in Heterogeneous Hybrid Views (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.