From Spectrum-free towards Baseline-view-free: Double-track Proximity Driven Multi-view Clustering

Authors: Shengju Yu, Zhibin Dong, Siwei Wang, Suyuan Liu, Ke Liang, Xinwang Liu, Yue Liu, Yi Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on multiple publicly available datasets confirm the effectiveness of proposed DTP-SF-BVF method. Concretely, we explicitly exploit the geometric properties between anchors via self-expression learning skill, and utilize topology learning strategy to feed captured anchoranchor features into anchor-sample graph so as to explore the manifold structure hidden within samples more adequately.
Researcher Affiliation Academia 1School of Computer, National University of Defense Technololgy, Hunan, China. 2Intelligent Game and Decision Lab, Beijing, China. EMAIL. Correspondence to: Siwei Wang <EMAIL>, Xinwang Liu <EMAIL>.
Pseudocode Yes Algorithm 1 The proposed DTP-SF-BVF Input: Multi-view data {Xp}v p=1, hyper-parameters λ, β. Output: Discrete cluster indicator matrix C. Initialize: {Ap, Tp, Bp, Sp}v p=1, C, α. 1: repeat 2: Update Ap via Eq. (5) 3: Update Tp via Eq. (7) 4: Update Bp via Eq. (9) 5: Update Sp via Eq. (11) 6: Update C via Eq. (14) 7: Update αp via Eq. (16) 8: until convergent
Open Source Code No The paper does not contain any explicit statement about code release or a link to a repository. Phrases like "We release our code..." or specific URLs are missing.
Open Datasets Yes Datasets We evaluate the algorithm performance on the following 7 datasets: DERMATO, CALTE7, Cora, REU7200, Reuters, CIF10Tra4, Fas MNI4V.
Dataset Splits No The paper mentions running experiments multiple times and calculating mean and variance, but it does not specify exact percentages, sample counts, or reference to predefined training/validation/test splits for the datasets used.
Hardware Specification No The paper does not mention any specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper does not provide specific names of software libraries or tools along with their version numbers, which are necessary for reproducibility.
Experiment Setup Yes Parameter Setup We search the hyper-parameters λ and β in [10 1, 100, 101, 102, 103] and [2 4, 2 2, 20, 22, 24] respectively. For all competitors, we download their source code and tune the parameters according to their provided guidelines. Three popular metrics are employed to measure the clustering results. For fairness, we run 20 times and calculate the mean and variance of clustering results.