Learning Cluster-Wise Anchors for Multi-View Clustering

Authors: Chao Zhang, Xiuyi Jia, Zechao Li, Chunlin Chen, Huaxiong Li

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive results demonstrate the effectiveness of our proposed method compared with some state-of-the-art MVC approaches.
Researcher Affiliation Academia 1Department of Control Science and Intelligence Engineering, Nanjing University 2School of Computer Science and Engineering, Nanjing University of Science and Technology
Pseudocode Yes Algorithm 1: CAMVC algorithm Input: Multi-view data {Xi}v i=1, cluster number k, parameters α, β, m. 1: Initialize Ai = 0, Hi = 0. 2: Construct the prior cluster indicator matrix Y. 3: while not converged do 4: Update Z by Eq. (7); 5: Update {Ai}v i=1 by Eq. (9); 6: Update {Hi}v i=1 by solving (11); 7: end while Output: Perform k-means on Z to obtain the clusters.
Open Source Code No For a fair comparison, we use the official codes of baselines. The paper does not provide concrete access to source code for the methodology described in this paper.
Open Datasets Yes We conduct experiments on seven popular datasets, including MSRC (Chen et al. 2021), BBCSport1, Wiki2, Notting Hill (Chen et al. 2021), Caltech1013, Fashion (Xiao, Rasul, and Vollgraf 2017), and MNIST4. Table 2 shows the general statistics of these datasets. (Footnotes provide URLs: 1http://mlg.ucd.ie/datasets/bbc.html 2http://www.svcl.ucsd.edu/projects/crossmodal/ 3https://paperswithcode.com/dataset/caltech-101 4http://yann.lecun.com/exdb/mnist/)
Dataset Splits No For our CAMVC, we tune α in the range of {10 3, 10 2, ..., 101}, β in {10 1, 10 2, ..., 103}, and m in {1, 3, 5}. ... The paper does not provide specific train/validation/test dataset splits, only mentions parameter tuning.
Hardware Specification Yes All experiments are conducted using MATLAB 2017b with i5-1230 CPU and 16GB RAM.
Software Dependencies Yes All experiments are conducted using MATLAB 2017b with i5-1230 CPU and 16GB RAM.
Experiment Setup Yes For our CAMVC, we tune α in the range of {10 3, 10 2, ..., 101}, β in {10 1, 10 2, ..., 103}, and m in {1, 3, 5}. ... following (Wang et al. 2022; Wan et al. 2023), we run 50 times k-means on final representation to report the best results. The optimal parameters of baselines are tuned by grid search in suggested ranges.