Improving the Generalization of Adversarial Training with Domain Adaptation
Authors: Chuanbiao Song, Kun He, Liwei Wang, John E. Hopcroft
ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical evaluations on Fashion-MNIST, SVHN, CIFAR-10 and CIFAR-100 demonstrate that ATDA can greatly improve the generalization of adversarial training and the smoothness of the learned models, and outperforms state-of-the-art methods on standard benchmark datasets. |
| Researcher Affiliation | Academia | Chuanbiao Song Department of Computer Science Huazhong University of Science and Technology Wuhan 430074, China cbsong@hust.edu.cn Kun He Department of Computer Science Huazhong University of Science and Technology Wuhan 430074, China brooklet60@hust.edu.cn Liwei Wang Department of Machine Intelligence Peking University wanglw@pku.edu.cn John E. Hopcroft Department of Computer Science Cornell University Ithaca 14850, NY, USA jeh@cs.cornell.edu |
| Pseudocode | Yes | Algorithm 1 Adversarial training with domain adaptation on network f(x) : Rd Rk. |
| Open Source Code | Yes | Code for these experiments is available at https: //github.com/JHL-HUST/ATDA. |
| Open Datasets | Yes | We consider four popular datasets, namely Fashion-MNIST (Xiao et al., 2017), SVHN (Netzer et al., 2011), CIFAR-10 and CIFAR-100 (Krizhevsky & Hinton, 2009). |
| Dataset Splits | No | The paper mentions training and testing data but does not specify explicit train/validation/test splits, percentages, or absolute counts for dataset partitioning. |
| Hardware Specification | Yes | All experiments are implemented on a single Titan X GPU. |
| Software Dependencies | No | The paper mentions optimizers (Adam) and components like ELU and Group Normalization, but it does not specify software names with version numbers (e.g., Python, TensorFlow, PyTorch versions) for reproducibility. |
| Experiment Setup | Yes | For all experiments, we set the hyper-parameter λ in Eq. (12) to 1/3 and the hyper-parameter α in Eq. (10) to 0.1. ... In the training phase, we use Adam optimizer with a learning rate of 0.001 and set the batch size to 64. ... For all adversarial training methods, the magnitude of perturbations is 0.1 in ℓ norm. |