Lyapunov-Stable Deep Equilibrium Models
Authors: Haoyu Chu, Shikui Wei, Ting Liu, Yao Zhao, Yuto Miyatake
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate Lya DEQ models under well-known adversarial attacks, and experimental results demonstrate significant improvement in robustness. |
| Researcher Affiliation | Academia | 1Institute of Information Science, Beijing Jiaotong University 2Graduate School of Information Science and Technology, Osaka University 3Beijing Key Laboratory of Advanced Information Science and Network Technology 4School of Computer Science, Northwestern Polytechnical University 5Cybermedia Center, Osaka University |
| Pseudocode | No | The paper provides architectural diagrams and mathematical formulations but does not include any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or link for the open-source code of the described methodology. |
| Open Datasets | Yes | We conduct a set of experiments on three standard datasets MNIST (Le Cun et al. 1998), CIFAR10/100 (Krizhevsky, Hinton et al. 2009), and SVHN (Netzer et al. 2011) |
| Dataset Splits | Yes | We conduct a set of experiments on three standard datasets MNIST (Le Cun et al. 1998), CIFAR10/100 (Krizhevsky, Hinton et al. 2009), and SVHN (Netzer et al. 2011) and The training epochs for MNIST, SVHN, and CIFAR are set to 10, 40, and 50. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments. |
| Software Dependencies | No | The paper only mentions 'Py Torch (Paszke et al. 2017) framework' without specifying a version number for PyTorch or any other software dependencies with their versions. |
| Experiment Setup | Yes | For optimization, we use Adam algorithm (Kingma and Ba 2014) with betas=(0.9, 0.999). We set the initial learning rate to 0.001 and set the learning rate of each parameter group using a cosine annealing schedule. The training epochs for MNIST, SVHN, and CIFAR are set to 10, 40, and 50. ... For both PGD and I-FGSM, the step size α is set to 1/255, and the number of steps n is calculated as n = min(ϵ 255+ 4, ϵ 255 1.25) . |