Self-Supervised Representation Learning with Meta Comprehensive Regularization
Authors: Huijie Guo, Ying Ba, Jie Hu, Lingyu Si, Wenwen Qiang, Lei Shi
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that our method achieves significant improvement in classification, object detection and instance segmentation tasks on multiple benchmark datasets. |
| Researcher Affiliation | Collaboration | 1Beihang University 2Institute of Software Chinese Academy of Sciences 3University of Chinese Academy of Sciences 4Gaoling School of Artificial Intelligence, Renmin University of China 5Beijing Key Laboratory of Big Data Management and Analysis Methods 6Meituan |
| Pseudocode | Yes | Algorithm 1: The main algorithm |
| Open Source Code | No | The paper does not include an explicit statement about releasing source code or provide a link to a code repository for the methodology described. |
| Open Datasets | Yes | For the classification task, we evaluate our proposed method on the following six image datasets, including CIFAR-10 and CIFAR-100 dataset (Krizhevsky 2009), STL-10 dataset (Coates, Ng, and Lee 2011), Tiny Image Net dataset (Le and Yang 2015), Image Net-100 dataset (Russakovsky et al. 2015), and Image Net dataset (Russakovsky et al. 2015). For transfer learning, we validate our method by the performance on the object detection and semantic segmentation tasks on COCO (Lin et al. 2014) dataset. |
| Dataset Splits | Yes | The detailed experimental setup follows the most common evaluation protocol for semi-supervised learning, as in Appendix B. ... We finetune the pretrained model using 1% and 10% training samples of Image Net following (Zbontar et al. 2021), and the top-1 and top-5 under linear evaluation are reported. |
| Hardware Specification | No | The paper does not specify the hardware used for running experiments, such as GPU models, CPU types, or cloud computing resources. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software components (e.g., programming languages, libraries, or frameworks) used in the experiments. |
| Experiment Setup | Yes | Default Setting. Each input sample generates two corresponding positive samples in the experiment. The image augmentation strategies comprise the following image transformations: random cropping, resizing, horizontal flipping, color jittering, converting to grayscale and gaussian blurring. Detailed experimental settings for different downstream tasks can be found in Appendix B. In the experiment, we use Resnet18 or Resnet50 as our base encoder network, along with a 3-layer MLP projection head to project the representation to a embedding space. |