Multimodal Graph Neural Architecture Search under Distribution Shifts
Authors: Jie Cai, Xin Wang, Haoyang Li, Ziwei Zhang, Wenwu Zhu
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we perform various experiments to verify the effectiveness of the proposed OMG-NAS method. Extensive experiments on real-world multimodal graph datasets demonstrate the superiority of our proposed method over SOTA baselines. |
| Researcher Affiliation | Academia | Jie Cai 1, Xin Wang 1,2*, Haoyang Li 1, Ziwei Zhang 1, Wenwu Zhu1,2* 1Department of Computer Science and Technology, Tsinghua University 2Beijing National Research Center for Information Science and Technology, Tsinghua University |
| Pseudocode | No | The paper describes the search algorithm in text and mathematical formulations (Section 3.4) but does not provide it in a structured pseudocode or algorithm block. |
| Open Source Code | No | The paper does not provide a direct link to source code or explicitly state that the code for the described methodology is open-source or available. |
| Open Datasets | No | The paper mentions 'Tencent dataset', 'Amazon review dataset', and 'Recipe dataset'. It indicates that data for Amazon and Recipe were extracted from general public websites (e.g., 'https://www.Amazon.com', 'https://www.simplyrecipes.com/'), but does not provide direct links, DOIs, specific repository names, or citations for the processed datasets used in the experiments. No source or access information is provided for the Tencent dataset. |
| Dataset Splits | No | The paper mentions using a 'validation dataset' for model selection and optimization (e.g., 'record its highest accuracy on the validation dataset'), but it does not specify the exact split percentages (e.g., 80/10/10) or sample counts for the training, validation, and test datasets. It describes OOD settings like domains, but not explicit data partitioning ratios. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running the experiments, such as GPU or CPU models, or memory specifications. |
| Software Dependencies | No | The paper mentions using 'pre-trained Bert' and 'pre-trained Vision Transformer (Vi T)' from 'open-source implementations (Wolf et al. 2020)', but it does not provide specific version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | For the Tencent dataset, we set the number of epochs to 200, the learning rate to 0.001, and the dimensions of the representations and hidden layers to 768 for both the text modality and visual modality. For the Amazon dataset, we set the number of epochs to 100, choose the learning rate from {0.001, 0.005, 0.01}, and set the dimensions of the representations and hidden layers to 128 for both the text modality and visual modalities. For the Recipe dataset, the number of epochs is set to be 50, the batch size is selected from {8,16,64}, and the learning rate is chosen from {0.001, 0.005, 0.01}. The dimensions of the representations and hidden layers are set to 200 for the text modality and 128 for the visual modality. The number of epochs for learning weights in MGFD is set to be 30 for all datasets. We utilize a two-layer MLP classifier. We report the mean values with standard deviations from 5 repeated experiments. |