COMIC: Multi-view Clustering Without Parameter Selection

Authors: Xi Peng, Zhenyu Huang, Jiancheng Lv, Hongyuan Zhu, Joey Tianyi Zhou

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We carry out experiments on five widely-used multi-view datasets comparing with nine state-of-the-art Mv C approaches in terms of two performance evaluation metrics.
Researcher Affiliation Collaboration Xi Peng 1 Zhenyu Huang 1 Jianchen Lv 1 Hongyuan Zhu 2 Joey Tianyi Zhou 3 1College of Computer Science, Sichuan University, Chengdu, China 2Institute for Infocomm Research, A*STAR, Singapore 3Institute of Performance Computing, A*STAR, Singapore.
Pseudocode Yes Algorithm 1 Cross-view Matching Clustering
Open Source Code No We use the sklearn code of k-means and SC, and implement our COMIC in python. Regarding to other seven tested method, we use the code released by the corresponding authors.
Open Datasets Yes We conduct experiments using five popular datasets, namely, Caltech101, Scene-15, Land Use-21, Still-DB, and MNISTUSPS. To be specific, Caltech101 (Li et al., 2015)... The used Scene-15 multi-view dataset (Zhang et al., 2017)... The used Land Use-21 dataset (Zhang et al., 2017)... The used Still-DB multi-view dataset (Zhang et al., 2017)... For the MNIST-USPS dataset...
Dataset Splits No The paper describes using five datasets but does not provide specific details on training, validation, or test splits (percentages, counts, or explicit standard split citations) to reproduce the data partitioning for their experiments.
Hardware Specification Yes All the experiments are implemented using MATLAB 2016a/Python 2.7 on a standard Linux Server with an Intel Xeon 2.10 GHz CPU and 32 GB RAM.
Software Dependencies Yes All the experiments are implemented using MATLAB 2016a/Python 2.7 on a standard Linux Server with an Intel Xeon 2.10 GHz CPU and 32 GB RAM.
Experiment Setup Yes Regarding to the precomputed similarity graph W(v), we set the neighbor size to 10 and adopt the cosine distance as the measurement. In brief, we tune the kernel width for SC with the parameter range of (0.001, 0.01, 0.1). For LRR, the value of λ ranges from 0.1 to 6.0 with an interval of 0.5.