Incomplete Contrastive Multi-View Clustering with High-Confidence Guiding

Authors: Guoqing Chao, Yi Jiang, Dianhui Chu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments compared with state-of-the-art approaches demonstrated the effectiveness and superiority of our method.
Researcher Affiliation Academia Guoqing Chao, Yi Jiang, Dianhui Chu Harbin Institute of Technology, 2 West Culture Road, Weihai, Shandong 264209, China guoqingchao10@gmail.com, jiangyijcx@163.com, chudh@hit.edu.cn
Pseudocode Yes Algorithm 1 Optimization of the proposed ICMVC
Open Source Code Yes Our code is publicly available at https://github.com/liunian-Jay/ICMVC.
Open Datasets Yes We used four commonly-used datasets in our experiments to evaluate our model. Scene-15: It consists of 4,485 images distributed in 15 scene categories with GIST and LBP features as two views. Land Use-21: It consists of 2100 satellite images from 21 categories with two views: PHOG and LBP. MSRC-V1: It is an image dataset consisting of 210 images in seven categories, including trees, buildings, airplanes, cows, faces, cars, and bicycles, with GIST and HOG features as two views. Noisy MNIST: the original images are used as view 1, and the sampled intra-class images with Gaussian white noise are used as view 2, and we use its subset containing 10k samples in the experiments.
Dataset Splits No The paper does not explicitly provide details about training/validation/test dataset splits. It mentions varying missing rates for evaluation but not how the datasets themselves were partitioned for training or validation purposes.
Hardware Specification Yes We implement ICMVC in Py Torch 1.12.1 and conduct all the experiments on Ubuntu 20.04 with NVIDIA 2080Ti GPU.
Software Dependencies Yes We implement ICMVC in Py Torch 1.12.1 and conduct all the experiments on Ubuntu 20.04 with NVIDIA 2080Ti GPU.
Experiment Setup Yes The Adam optimizer is adopted, and the learning rate is set to 0.001, the hyper-parameter K is set to 10. The instance-level temperature parameter τI is fixed at 1.0, and the cluster-level parameter τC is fixed at 0.5. We observe that it can fully converges after 500 epoches after training the network, thus 500 epoches is set to terminate.