Robust Meta-learning with Sampling Noise and Label Noise via Eigen-Reptile

Authors: Dong Chen, Lingfei Wu, Siliang Tang, Xiao Yun, Bo Long, Yueting Zhuang

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that Eigen-Reptile significantly outperforms the baseline, Reptile, by 22.93% and 5.85% on the corrupted and clean dataset, respectively. 5. Experimental Results and Discussion
Researcher Affiliation Collaboration 1College of Computer Science and Technology, Zhe Jiang University, Hang Zhou, China 2JD.COM Silicon Valley Research Center, 675 E Middlefield Rd, Mountain View, CA 94043 USA.
Pseudocode Yes Algorithm 1 Eigen-Reptile
Open Source Code Yes The code and data for the proposed method are provided for research purposes https://github.com/Anfeather/Eigen-Reptile.
Open Datasets Yes We verify the effectiveness of Eigen-Reptile alleviate overfitting sampling noise on two clean few-shot classification datasets Mini-Imagenet (Vinyals et al., 2016) and CIFAR-FS (Bertinetto et al., 2018). The Mini-Imagenet dataset contains 100 classes, each with 600 images. We follow (Ravi & Larochelle, 2016) to divide the dataset into three disjoint subsets: meta-training set, meta-validation set, and meta-testing set with 64 classes, 16 classes, and 20 classes, respectively.
Dataset Splits Yes We follow (Ravi & Larochelle, 2016) to divide the dataset into three disjoint subsets: meta-training set, meta-validation set, and meta-testing set with 64 classes, 16 classes, and 20 classes, respectively.
Hardware Specification Yes All experiments run on a 2080 Ti.
Software Dependencies No The paper mentions software like Adam and PyTorch but does not specify their version numbers or other software dependencies with version details.
Experiment Setup Yes All meta-learners use the same regressor that is trained for 30000 iterations with inner loop steps 5, batch size 10, and a fixed inner loop learning rate of 0.02. Our model is trained for 100000 iterations with a fixed inner loop learning rate of 0.0005, 7 inner-loop steps and batch size 10.