Incomplete Multi-View Multi-Label Classification via Diffusion-Guided Redundancy Removal
Authors: Shilong Ou, Zhe Xue, Lixiong Qin, Yawen Li, Meiyu Liang, Junjiang Wu, Xuyun Zhang, Amin Beheshti, Yuankai Qi
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on five datasets, and the proposed method achieves favorable performance against several state-of-the-art methods on the multi-view multi-label classification task. Comprehensive experiments validate the superior performance of our proposed Diff Summ framework. Extensive experiments show the favorable performance on five datasets compared to state-of-the-art methods. |
| Researcher Affiliation | Academia | 1Beijing University of Posts and Telecommunications, China 2Macquarie University, Australia |
| Pseudocode | No | The paper describes the methodology using textual explanations and mathematical equations, but it does not contain a clearly labeled pseudocode or algorithm block. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing source code, nor does it provide a link to a code repository. |
| Open Datasets | Yes | Our research employs the methodology suggested by (Tan et al. 2018) to assess the effectiveness of Diff Summ. The datasets included in this evaluation are as follows: Corel 5k (Duygulu et al. 2002)... Pascal07 (Everingham et al. 2010)... Espgame (von Ahn and Dabbish 2004)... IAPRTC12 (Grubinger et al. 2006)... Mirflickr (Huiskes and Lew 2008)... |
| Dataset Splits | Yes | The results of various methods on five datasets with 70% training samples, 50% missing views and missing labels. In terms of model architecture... To simulate a scenario of incompleteness, we randomly remove 50% of the views from the data in each dataset. Within every category, we randomly mark 50% of both the positive and negative tags as unavailable. We set the masking rate kmask to 0.25 across all experiments. Figure 2 illustrates the model s performance under various view and label missing ratios. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU or CPU models used for the experiments. |
| Software Dependencies | No | The paper describes model architectures (e.g., "2 multilayer perceptrons", "UNet", "cross-attention mechanism") but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | In terms of model architecture, we employ 2 multilayer perceptrons with four fully connected layers as the encoder and decoder for the autoencoder. For the diffusion model, we utilize a UNet (Ronneberger, Fischer, and Brox 2015) with a cross-attention mechanism, akin to that described in (Rink et al. 2021), comprising five convolutional layers. ... We set the masking rate kmask to 0.25 across all experiments. To validate the reliability of our findings, we conduct each experiment 10 times and calculate the average and variance of the results. For ksig, we design various methods for evaluation, namely fixed value, cumulative addition, and cumulative multiplication. The fixed value method sets ksig to 1 for rounds 1-3, 2 for rounds 4-6, and 3 for rounds beyond 6; cumulative addition is tested with ksig starting at 1 and increasing by 0.25, 0.5, 1, in each round respectively; cumulative multiplication started with ksig at 1, with each round multiplying by 1.25, 1.5, 2, respectively. |