Unified K-Means Clustering with Label-Guided Manifold Learning

Authors: Qianqian Wang, Mengping Jiang, Zhengming Ding, Quanxue Gao

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

Reproducibility Variable Result LLM Response
Research Type Experimental The detailed experimental results demonstrate the efficacy of our proposed methodology. In this chapter, we conducted relevant experiments on two toy datasets and ten benchmark datasets, and selected 6 classic clustering comparison algorithms for comparison.
Researcher Affiliation Academia 1School of Communication Engineering, Xidian University, Xi an, China. 2Department of Computer Science, Tulane University, New Orleans, LA. Correspondence to: Quanxue Gao <EMAIL>.
Pseudocode Yes The comprehensive flow of the algorithm can be found in Algorithms 1 and 2. Algorithm 1 Optimizing F Algorithm 2 Solving problem (24)
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 Datasets: We selected the following ten datasets: CMUPIE (Sim et al., 2002), digits (Kusetogullari et al., 2020), FERET (Phillips et al., 2000), Mpeg7 (Bober, 2001), olivetti (Samaria & Harter, 1994), Palm1, Pengdigits (Liu & Wechsler, 1997), PEAL (Wang & Tang, 2004), STL10 (Coates et al., 2011), and USPS (Hull, 1994). Detailed information on the datasets is provided in the Appendix A.
Dataset Splits No The paper mentions dividing toy datasets into clusters (e.g., "There are a total of four hundred samples, divided into two clusters, with 200 samples in each cluster" for the Two-spiral Dataset). However, it does not provide specific training, validation, or test splits for any of the datasets used in the experiments.
Hardware Specification Yes Our experiments were conducted on a Windows 11 system, 13th Gen Intel(R) Core(TM) CPU, and MATLAB R2023a.
Software Dependencies Yes Our experiments were conducted on a Windows 11 system, 13th Gen Intel(R) Core(TM) CPU, and MATLAB R2023a.
Experiment Setup Yes A.4. Parameter setting Since Equation (24) contains the parameter λ, we need to set the parameter λ for all four distances. In addition, Our-KNN needs to set the value of the nearest neighbor K; using the Gaussian kernel function K(x, y) = exp(−kx − yk2 /σ2 ), Our-K-ED contains the parameter σ. ... Our-ED: The values of λ for CMUPIE, digits, FERET, Mpeg7, olivetti, Palm, Pengdigits, PEAL, STL10, and USPS are 7696000, 2886000, 1000, 1200, 1801000, 13126000, 500000, 88000, 500000, 52600.