Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Rethinking Guidance Information to Utilize Unlabeled Samples: A Label Encoding Perspective
Authors: Yulong Zhang, Yuan Yao, Shuhao Chen, Pengrong Jin, Yu Zhang, Jian Jin, Jiangang Lu
ICML 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, we verify the superiority of the LERM under several label insufficient scenarios. The codes are available at https://github.com/zhangyl660/LERM. ... 5. Experiments We assess the performance of the LERM on three typical label insufficient scenarios, including SSL, UDA, and SHDA. The loss function defined in Eq. (5) adopts the โ1 distance, i.e., L(mu c , ec) = mu c ec 1, where 1 denotes the โ1 norm of a vector. For comparisons among different loss functions, please refer to Appendix E.2. |
| Researcher Affiliation | Collaboration | 1Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen, China 2College of Control Science and Engineering, Zhejiang University, Hangzhou, China 3Beijing Teleinfo Technology Company Ltd., China Academy of Information and Communications Technology, Beijing, China 4Research Institute of Industrial Internet of Things, China Academy of Information and Communications Technology, Beijing, China. |
| Pseudocode | No | The paper describes methods but does not include any explicit pseudocode blocks or algorithm listings. |
| Open Source Code | Yes | The codes are available at https://github.com/zhangyl660/LERM. |
| Open Datasets | Yes | We evaluate the LERM on four SSL benchmark datasets, including CIFAR-10 (Krizhevsky et al., 2009), CIFAR-100 (Krizhevsky et al., 2009), DTD (Cimpoi et al., 2014), and Image Net-1K (Deng et al., 2009). ... We evaluate the LERM on three UDA benchmark datasets, i.e., Office-31 (Saenko et al., 2010), Office-Home (Venkateswara et al., 2017), Vis DA-2017 (Peng et al., 2017), and Image Net. |
| Dataset Splits | No | We conduct experiments on the SSL tasks with limited labeled samples. ... We set the batch sizes of both domains to 32. While the paper uses standard datasets, it does not explicitly provide the specific percentages or counts for a train/validation split used in their experiments, nor does it specify how a validation set is used from their limited labeled data setup. |
| Hardware Specification | Yes | The experiments on SSL and UDA tasks are conducted on a NVIDIA V100 GPU, and the experiments in SHDA tasks are conducted on a NVIDIA 3090 GPU. |
| Software Dependencies | No | The paper mentions optimizers like SGD and Adam, and deep learning architectures like ResNet-18 and ResNet-50, but does not provide specific version numbers for software libraries or programming languages (e.g., Python, PyTorch, TensorFlow). |
| Experiment Setup | Yes | The parameter ฮป in Eq. (9) for SSL tasks is shown in Table 8, and the parameter ยต in Eq. (9) is set to 0.1. We use mini-batch stochastic gradient descent (SGD) with a momentum of 0.9 as the optimizer, and the batch sizes of labeled and unlabeled samples are both set to 32 on the CIFAR-10, CIFAR-100, and DTD datasets and 512 on the Image Net dataset. |