Towards Escaping from Class Dependency Modeling for Multi-Dimensional Classification
Authors: Teng Huang, Bin-Bin Jia, Min-Ling Zhang
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on benchmark data sets demonstrate that DCOM outperforms other stateof-the-art MDC approaches. ... Firstly, related works are briefly reviewed in Section 2. Secondly, the details of the proposed DCOM approach are presented in Section 3. Thirdly, experimental results of comparative studies on benchmark multi-dimensional data sets are reported in Section 4. Finally, Section 5 concludes this paper. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, Southeast University, Nanjing 210096, China 2Key Lab. of Computer Network and Information Integration (Southeast University), MOE, China 3College of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou 730050, China. Correspondence to: Min-Ling Zhang <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 presents the pseudo code of the proposed DCOM approach. Algorithm 1 The DCOM approach |
| Open Source Code | No | The paper does not contain any explicit statement about providing source code or a link to a code repository for the described methodology. It only mentions URLs for datasets. |
| Open Datasets | Yes | In this paper, we use seventeen real-world MDC data sets for experimental studies, including fourteen structured data sets4 and three unstructured data sets: BP4D (Zhang et al., 2013; 2014),5 Deep Fashion (Liu et al., 2016)6 and SEWA (Kossaifi et al., 2021).7 Table 1 summarizes basic characteristics, including the number of examples (#Exam.), the number of dimensions (#Dim.), the number of labels in each dimension (#Labels/Dim.) and the number of features (#Feat.). ... 4https://palm.seu.edu.cn/zhangml/ Resources.htm#MDC_data ... 5https://www.cs.binghamton.edu/ lijun/Research/3DFE/3DFE_Analysis.html. ... 6https://mmlab.ie.cuhk.edu.hk/projects/ DeepFashion/AttributePrediction.html. ... 7https://db.sewaproject.eu/. |
| Dataset Splits | Yes | For structured datasets, tenfold cross validation are conducted where the mean metric value as well as the standard derivation are recorded for comparison. |
| Hardware Specification | No | The paper mentions that ResNet-18 was used as a feature extractor, but it does not provide any specific details about the hardware (e.g., GPU model, CPU type, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions using SGD for network optimization and ReLU as an activation function, and ResNet-18 as a feature extractor. However, it does not specify any version numbers for these software components or for any other libraries/frameworks used, which is required for reproducibility. |
| Experiment Setup | Yes | For all neural networks G, R, T and {Hj|1 j q}, we employ the multi-layer perceptron (MLP) with one hidden layer, configured with hidden dimension of 512. The dimensionality of latent variable Z is set as 512. In the overall loss function, i.e., Eq.(17), the trade-off parameter α and β are both set as 1 (please refer to detailed discussions on parameter sensitivities in Section 4.3.2). All activation functions are fixed as ReLU followed by a dropout layer (Srivastava et al., 2014) with dropping probability of 0.5. For network optimization, we utilize SGD with a batch size of 512, momentum of 0.9 and weight decay of 10-4. We only adopt experimental results of the last epoch and the number of epoch is uniformly set as 500 for all data sets. |