Self-supervised Trusted Contrastive Multi-view Clustering with Uncertainty Refined

Authors: Shizhe Hu, Binyan Tian, Weibo Liu, Yangdong Ye

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on five datasets show that the proposed method has significant improvements in effectiveness compared with the latest methods.
Researcher Affiliation Academia Shizhe Hu, Binyan Tian, Weibo Liu, Yangdong Ye * School of Computer Science and Artificial Intelligence, Zhengzhou University, Zhengzhou, China EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Algorithm for STCMC-UR
Open Source Code Yes Code https://github.com/Shizhe Hu.
Open Datasets Yes Datasets. Five frequently-used datasets are selected in the experiments shown in Table 1, and a detailed introduction to them are: Caltech1 image dataset has two multi-view versions: Caltech-2V is categorized into 7 classes with a total of 1400 images. It contains two kinds of features. Caltech3V includes the same classes and images compared to Caltech-2V, but introduces an additional feature. COIL202 dataset contains grayscale images categorized into 20 clusters. Each object is photographed at 5-degree intervals, resulting in 72 images per object. Event83 contains 8 sports event categories with 1,579 samples and three features. It poses challenges due to significant background variability 1https://data.caltech.edu/records/mzrjq-6wc02 2http://www1.cs.columbia.edu/CAVE/software/softlib/coil100.php 3http://vision.stanford.edu/lijiali/event dataset/ NUS224 consists of 10,155 images across 22 classes from the original NUS-WIDEObject dataset, in which two kinds of features are adopted. 4https://lms.comp.nus.edu.sg/wp-content/uploads/2019/ research/nuswide/NUS-WIDE.html
Dataset Splits No The paper mentions using datasets but does not explicitly describe the splits for training, validation, or testing, nor does it refer to standard predefined splits with a citation. It mentions 'training batch size' but not how the overall data was partitioned.
Hardware Specification Yes We ran 20 times and each for 100 epochs on PyTorch 1.13.0 platform (Python 3.8) equipped with a 24GB NVIDIA RTX-4090D GPU on Windows 10 system.
Software Dependencies Yes We ran 20 times and each for 100 epochs on Py Torch 1.13.0 platform (Python 3.8) equipped with a 24GB NVIDIA RTX-4090D GPU on Windows 10 system.
Experiment Setup Yes Implement Details. We ran 20 times and each for 100 epochs on Py Torch 1.13.0 platform (Python 3.8) equipped with a 24GB NVIDIA RTX-4090D GPU on Windows 10 system. We used two widely applied metrics for evaluation: ACC (Accuracy) and NMI (Normalized Mutual Information). Higher values for both metrics indicate better performance. For the datasets, the training batch size was consistently set as 100. The Adam optimizer was adopted with the learning rate of 0.001.