Boosting Active Learning via Improving Test Performance
Authors: Tianyang Wang, Xingjian Li, Pengkun Yang, Guosheng Hu, Xiangrui Zeng, Siyu Huang, Cheng-Zhong Xu, Min Xu8566-8574
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our method on classical image classification and semantic segmentation tasks. To demonstrate its competency in domain applications and its robustness to noise, we also validate our method on a cellular imaging analysis task, namely cryo-Electron Tomography subtomogram classification. Results demonstrate that our method achieves superior performance against the state of the art. |
| Researcher Affiliation | Collaboration | 1Austin Peay State University, 2Baidu Inc, 3Tsinghua University, 4Oosto, 5Harvard University, 6University of Macau, 7Carnegie Mellon University |
| Pseudocode | Yes | Algorithm 1: Proposed Active Learning Framework. |
| Open Source Code | No | We refer readers to https://arxiv.org/pdf/2112.05683.pdf for the full version of this paper which includes the appendix and source code link. |
| Open Datasets | Yes | We exploit five widely used image classification datasets in our experiments, namely Cifar10 (Krizhevsky, Hinton et al. 2009), Cifar100 (Krizhevsky, Hinton et al. 2009), SVHN (Netzer et al. 2011), Caltech101 (Fei-Fei, Fergus, and Perona 2006), and Image Net (Deng et al. 2009). |
| Dataset Splits | Yes | Both Cifar datasets include 50000 training and 10000 test samples, distributed across 10 and 100 classes, respectively. SVHN also consists of 10 classes, while it includes more samples than the two Cifar datasets, namely 73257 for training and 26032 for testing. For Image Net, we conduct the experiments with 5 AL cycles (i.e. annotation budget varies from 10% to 30%) and the results are averaged over 2 runs, which are acceptable for very large-scale datasets. We follow the common practice to report model performance on the validation set that consists of 50000 samples. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU models, CPU types, or memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers, such as Python version, or library versions like PyTorch or TensorFlow. |
| Experiment Setup | No | The paper describes the number of AL cycles and annotation budgets, but does not provide specific training hyperparameters such as learning rate, batch size, or optimizer details in the main text, instead referring to Appendix A.8 for training details. |