Memorization Weights for Instance Reweighting in Adversarial Training

Authors: Jianfu Zhang, Yan Hong, Qibin Zhao

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To evaluate the effectiveness of Meo W, we conducted extensive experiments on three datasets, including CIFAR10 (Krizhevsky et al. 2009), CIFAR-100 (Krizhevsky et al. 2009), and SVHN (Netzer et al. 2011). We choose popular Pre Act Res Net (He et al. 2016) and Wide Res Net (Zagoruyko and Komodakis 2016) as backbones in our experiments.
Researcher Affiliation Academia Jianfu Zhang1, Yan Hong2, and Qibin Zhao1 * 1RIKEN AIP 2Shanghai Jiao Tong University
Pseudocode Yes Please refer to the Appendix for the pseudocodes of the whole algorithm.
Open Source Code No The paper mentions pseudocodes in the Appendix but does not provide an explicit statement or link for open-source code for the methodology.
Open Datasets Yes To evaluate the effectiveness of Meo W, we conducted extensive experiments on three datasets, including CIFAR10 (Krizhevsky et al. 2009), CIFAR-100 (Krizhevsky et al. 2009), and SVHN (Netzer et al. 2011).
Dataset Splits No The paper mentions 'test robustness' and 'training process' but does not explicitly provide specific details about train/validation/test dataset splits (e.g., percentages, sample counts, or predefined split citations) used for reproducibility.
Hardware Specification Yes All training process is conducted on n Vidia A100 GPU.
Software Dependencies No The paper describes training parameters and methods but does not list specific software dependencies (e.g., Python, PyTorch, TensorFlow) with their version numbers.
Experiment Setup Yes Training Parameters We train the networks batch stochastic gradient descent with momentum 0.9, weight decay 5 10 4. The training process takes 200 epochs. Batch size is 128, and the initial learning rate is 0.1. Learning rate is divided by 10 after the 100-th and 150-th epoch. [...] Hyperparameters For the proposed Meo W, we set the size of the codebook N = 2048 and the temperature T = 10 in Eqn. (4). The trade-off parameter for AT α in Eqn. (5) was set to 0.2 and for TRADES λ in Eqn. (6) it was 6.