Masked Two-channel Decoupling Framework for Incomplete Multi-view Weak Multi-label Learning

Authors: Chengliang Liu, Jie Wen, Yabo Liu, Chao Huang, Zhihao Wu, Xiaoling Luo, Yong Xu

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We have conducted sufficient and convincing experiments to confirm the effectiveness and advancement of our model.
Researcher Affiliation Academia 1School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen 2School of Cyber Science and Technology, Sun Yat-sen University (Shenzhen Campus) 3College of Computer Science and Software Engineering, Shenzhen University
Pseudocode Yes Algorithm 1 Training process of MTD
Open Source Code No The code can be found at https://justsmart.github.io/. This appears to be a personal homepage rather than a direct link to the source code repository for the specific methodology described in the paper.
Open Datasets Yes Consistent with existing i Mv WMLC methods [26, 25, 20], we adopt five famous multi-view multilabel datasets2 to validate our model, i.e., Corel5k [32], Pascal07 [33], ESPGame [34], IAPRTC12 [35], and MIRFLICKR [36]. http://lear.inrialpes.fr/people/guillaumin/data.php
Dataset Splits No 70% of samples with missing views and missing labels are randomly selected as the training set. The paper specifies a training set percentage but does not explicitly detail separate validation and test splits.
Hardware Specification Yes Our TMD is implemented by Pytorch and Mind Spore frameworks on Ubuntu operating system with a single RTX 3090 GPU and an i7-12900k CPU.
Software Dependencies No Our TMD is implemented by Pytorch and Mind Spore frameworks. Specific version numbers for these frameworks are not provided.
Experiment Setup Yes The learning rate is set to 0.1 and the Stochastic Gradient Descent (SGD) optimizer is chosen for training model. The batch size and momentum are 128 and 0.9 for all five datasets. ... we uniformly set α = 0.4 and β = 0.4 in our experiments. For γ, our method is not sensitive to it, so we set γ = 0.1 for all five datasets. ... we uniformly set the mask rate σ = l/dv to 0.25 for all datasets... For DICNet and MTD, we report the time spent in 100 epochs.