Quality-Aware Self-Training on Differentiable Synthesis of Rare Relational Data
Authors: Chongsheng Zhang, Yaxin Hou, Ke Chen, Shuang Cao, Gaojuan Fan, Ji Liu
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on 20 benchmark datasets of different domains, including 14 industrial datasets. The results show that our method significantly outperforms state-of-the-art methods, including two recent GAN-based data synthesis schemes. |
| Researcher Affiliation | Collaboration | Chongsheng Zhang1, Yaxin Hou1, Ke Chen2,3,*, Shuang Cao1, Gaojuan Fan1, Ji Liu4 1Henan University 2South China University of Technology 3Peng Cheng Laboratory 4Baidu Research |
| Pseudocode | No | The paper describes the proposed method in detail but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Codes are available at https://github.com/yaxinhou/QAST. |
| Open Datasets | Yes | We evaluate comparative methods on 6 multi-class imbalanced datasets from the UCI repository. Moreover, we include 14 industrial datasets, which are the CWRU Bearing (Li et al. 2020) and Gearbox fault diagnosis datasets (Lin and Zuo 2003), and the NASA software defect repository that contains 12 datasets (Shepperd et al. 2013). |
| Dataset Splits | Yes | In the experiments, 10 cross-validations are adopted. Since there are no official splits for the above datasets, in each validation we randomly select half of the samples of each dataset as the training set, and use the other half as the test set. |
| Hardware Specification | No | The paper does not provide specific hardware details such as CPU/GPU models, memory, or cloud instance types used for experiments. |
| Software Dependencies | No | The paper mentions software components like 'scikit-learn library' and 'Adam optimizer' but does not specify version numbers for any of its software dependencies. |
| Experiment Setup | Yes | In all the experiments, the Adam optimizer is used with a default learning rate of 0.0002, and the training epochs is set to 300 for all different methods. |