Max-Mahalanobis Anchors Guidance for Multi-View Clustering
Authors: Pei Zhang, Yuangang Pan, Siwei Wang, Shengju Yu, Huiying Xu, En Zhu, Xinwang Liu, Ivor Tsang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Experiment This section compares MAGIC with state-of-the-art methods. We first introduce the datasets, compared methods and the experimental setup, followed by the detailed analysis. 4.1 Experimental Setup We conduct experiments on ten widely-used datasets: BBC, Wikipedia, Reuters, 100Leaves, Cora, Wiki fea, ALOI-100, VGGFace, You Tube Face, CIFAR100, denoted as Ds1 Ds10 in following figures for simplicity. Detailed information is presented in the appendix. We compare our approach with nine methods that encompass both fixed-anchor-based MVC methods... We adopt the officially released codes for fair comparison... We utilized widely used clustering metrics such as Accuracy (ACC), Normalized Mutual Information (NMI), Purity and Fscore, where higher values indicate better performance. |
| Researcher Affiliation | Academia | 1 College of Computer Science and Technology, National University of Defense Technology, Changsha, 410073, China 2 Centre for Frontier AI Research, Agency for Science, Technology and Research, 138632, Singapore 3 Institute of High Performance Computing, Agency for Science, Technology and Research, 138632, Singapore 4 Intelligent Game and Decision Lab, Beijing, 100091, China 5 School of Computer Science and Technology, Zhejiang Normal University, Jinhua, 321004, China EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: MMA guidance for multi-view clustering Input: Multi-view data {X(i)}V i=1, constant C, clusters K. Initialize: Initialize B by concatenating the identity matrix and the zero matrix. Initialize γi with the average weight 1 V . 1: Generate Max-Mahalanobis Anchors µ by Eq. (4). 2: while not converged do 3: Update P(i) = Um V m. 4: Update B by solving problem Eq. (6). 5: Update γi = 1 βi PV i=1 1 βi . 6: end while Output: Perform k-means on the left singular vector Ub of B to obtain the final clustering results. |
| Open Source Code | No | The text does not contain any explicit statement about the release of source code for the methodology described in this paper, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We conduct experiments on ten widely-used datasets: BBC, Wikipedia, Reuters, 100Leaves, Cora, Wiki fea, ALOI-100, VGGFace, You Tube Face, CIFAR100, denoted as Ds1 Ds10 in following figures for simplicity. |
| Dataset Splits | No | The paper mentions using several datasets but does not provide specific details on training/test/validation splits, percentages, or methodology for data partitioning. |
| Hardware Specification | Yes | All experiments are conducted using MATLAB 2023b on the system equipped with an AMD EPYC 7513 32-Core Processor and 64GB of memory. |
| Software Dependencies | Yes | All experiments are conducted using MATLAB 2023b on the system equipped with an AMD EPYC 7513 32-Core Processor and 64GB of memory. |
| Experiment Setup | Yes | Optimal parameters are determined via grid search within established ranges. The constant C was set to 1, simplifying hyperparameter tuning to only balance parameter λ, which is explored within {0.01, 1, 10, 100, 1000} based on previous studies. |