Robust Contrastive Multi-view Clustering against Dual Noisy Correspondence

Authors: Ruiming Guo, Mouxing Yang, Yijie Lin, Xi Peng, Peng Hu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on five widely-used multi-view benchmarks, in comparison with eight competitive multi-view clustering methods, verify the effectiveness of our method in addressing the DNC problem.
Researcher Affiliation Academia Ruiming Guo1 , Mouxing Yang1 , Yijie Lin1, Xi Peng1,2, Peng Hu1 1College of Computer Science, Sichuan University, China 2State Key Laboratory of Hydraulics and Mountain River Engineering, Sichuan University, China
Pseudocode No The paper does not contain a pseudocode block or clearly labeled algorithm.
Open Source Code Yes The code is available at https://github.com/XLearning-SCU/2024-NeurIPS-CANDY.
Open Datasets Yes The experiments are carried out on the following five widely-used multi-view learning datasets. Scene-15 [42]... Caltech-101 [43]... Land Use-21 [45]... Reuters [47]... NUS-WIDE [49].
Dataset Splits No The paper does not specify explicit train/validation/test dataset splits. It states: 'Since Mv C requires training and clustering on the same dataset, we conduct the view realignment strategy on the learned representation by following the PVP studies [20, 21].'
Hardware Specification Yes All evaluations are conducted on Ubuntu 20.04 OS with NVIDIA 3090 GPUs.
Software Dependencies Yes In the experiment, CANDY is implemented with PyTorch 2.1.2
Experiment Setup Yes the model is optimized with the Adam [41] optimizer with a learning rate of 0.002 across all experiments, with a batch size fixed to 1024. ... The scale parameter σ in Eq. 3 is fixed as 0.07 across all experiments. ... η being a denoising hyper-parameter fixed as 0.2 in our experiments. ... λ is fixed as 0.2 in our experiments. ... for the first 20 epochs of training.