Collaborative Similarity Fusion and Consistency Recovery for Incomplete Multi-view Clustering

Authors: Bingbing Jiang, Chenglong Zhang, Xinyan Liang, Peng Zhou, Jie Yang, Xingyu Wu, Junyi Guan, Weiping Ding, Weiguo Sheng

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the effectiveness and superiority of SFCR.
Researcher Affiliation Academia 1Hangzhou Normal University, Hangzhou, China 2Shanxi University, Taiyuan, China 3Anhui University, Hefei, China 4University of Technology Sydney, NSW, Australia 5Hong Kong Polytechnic University, Hong Kong SAR, China 6Nantong University, Nantong, China
Pseudocode Yes Algorithm 1: Optimization procedures for SFCR
Open Source Code No The paper does not contain any explicit statements about releasing source code or provide links to a code repository.
Open Datasets Yes six real-world datasets are employed, including 3sources (Jiang et al. 2021), Webkb (Wang, Yang, and Liu 2020), MSRC-v1 (Wang, Yang, and Liu 2020), ORL (Zhang et al. 2023b), Leaves (Zhang et al. 2024a) and COIL100 (Zhang et al. 2024b)
Dataset Splits Yes the incomplete multi-view data are generated by randomly choosing samples from the complete datasets. Specifically, we vary the incomplete ratio on each view from the range of {0%, 20%, 30%}, while ensuring every sample appears at least once.
Hardware Specification Yes All experiments are conducted in Matlab R2022a on Windows 10 with a 3.2GHz CPU and 32GB RAM.
Software Dependencies Yes All experiments are conducted in Matlab R2022a on Windows 10
Experiment Setup Yes In SFCR, β can be dynamically updated based on the number of connection components, thus we initialize β = 1 and search λ in the grid of {10 3, 10 2, , 103}. Besides, since other methods require the post-precessing to obtain the clustering labels, such that we repeated the Kmeans 20 times to achieve the average clustering results.