Multi-Label Supervised Contrastive Learning
Authors: Pingyue Zhang, Mengyue Wu
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate, we conduct direct classification and transfer learning on several multi-label datasets, including widely-used image datasets such as MS-COCO and NUS-WIDE. Validation indicates that our method outperforms the traditional multilabel classification method and shows a competitive performance when comparing to other existing approaches. |
| Researcher Affiliation | Academia | Mo E Key Lab of Artificial Intelligence, AI Institute X-LANCE Lab, Department of Computer Science and Engineering Shanghai Jiao Tong University, Shanghai, China {williamzhangsjtu, mengyuewu}@sjtu.edu.cn |
| Pseudocode | No | The paper provides mathematical formulations for its loss function and describes its framework, but it does not include a block explicitly labeled as "Pseudocode" or "Algorithm". |
| Open Source Code | No | The paper does not contain an explicit statement or a link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | We evaluate our method by measuring the multi-label classification performance on several image datasets, including MS-COCO (Lin et al. 2014), NUS-WIDE (Chua et al. 2009), Objects365 (Shao et al. 2019), MIRFLICKR (Huiskes and Lew 2008), and PASCAL-VOC (VOC2007) (Everingham et al. 2007). |
| Dataset Splits | Yes | The dataset is divided into training, validation, and testing sets. |
| Hardware Specification | No | The paper mentions, "Experiments have been carried out on the PI supercomputer at Shanghai Jiao Tong University." This is a general statement about the computing environment but does not provide specific hardware details such as GPU/CPU models or memory specifications. |
| Software Dependencies | No | The paper mentions software components like "Res Net-50" (an architecture), "SGD optimizer", "Adam optimizer", "MLP", "Re LU activation function", and "BCE Loss". However, it does not provide specific version numbers for any libraries, frameworks (e.g., PyTorch, TensorFlow), or programming languages used. |
| Experiment Setup | Yes | For image datasets, during the pretraining stage, we train the model for 400 epochs using SGD optimizer with an initial learning rate of 0.1 and a cosine learning rate scheduler. A batch size of 64 is utilized for all datasets except for the Objects365 dataset, where a batch size of 128 is employed. For vector datasets, the model is trained for 150 epochs using the Adam optimizer with an initial learning rate of 0.0004, coupled with a cosine learning rate scheduler. A batch size of 256 is utilized, except for the yeast and scene datasets, where a batch size of 32 is used. |