Parseval Networks: Improving Robustness to Adversarial Examples
Authors: Moustapha Cisse, Piotr Bojanowski, Edouard Grave, Yann Dauphin, Nicolas Usunier
ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that Parseval networks match the state-of-the-art in terms of accuracy on CIFAR-10/100 and Street View House Numbers (SVHN), while being more robust than their vanilla counterpart against adversarial examples. |
| Researcher Affiliation | Industry | Moustapha Cisse 1 Piotr Bojanowski 1 Edouard Grave 1 Yann Dauphin 1 Nicolas Usunier 1 1Facebook AI Research. Correspondence to: Moustapha Cisse <moustaphacisse@fb.com>. |
| Pseudocode | Yes | Algorithm 1 Parseval Training |
| Open Source Code | No | The paper does not provide any explicit statement about making the source code available, nor does it include a link to a code repository. |
| Open Datasets | Yes | We evaluate the effectiveness of Parseval networks on well-established image classification benchmark datasets namely MNIST, CIFAR-10, CIFAR-100 (Krizhevsky, 2009) and Street View House Numbers (SVHN) (Netzer et al.). |
| Dataset Splits | Yes | Each of the CIFAR datasets is composed of 60K natural scene color images of size 32 32 split between 50K training images and 10K test images. ... we initially use 5K images from the training as a validation set. ... For SVHN...we randomly sample 10000 images from the available extra set of about 600K images as a validation set and combine the rest of the pictures with the official training set. |
| Hardware Specification | No | The paper mentions 'modern GPU architecture' but does not provide specific details on the CPU, GPU models, or memory used for experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers, such as programming languages, libraries, or specific deep learning frameworks. |
| Experiment Setup | Yes | We train the networks with stochastic gradient descent using a momentum of 0.9. On CIFAR datasets, the initial learning rate is set to 0.1 and scaled by a factor of 0.2 after epochs 60, 120 and 160, for a total number of 200 epochs. We used mini-batches of size 128. For SVHN, we trained the models with mini-batches of size 128 for 160 epochs starting with a learning rate of 0.01 and decreasing it by a factor of 0.1 at epochs 80 and 120. ... The dropout rate use is 0.3 for CIFAR and 0.4 for SVHN. For Parseval regularized models, we choose the value of the retraction parameter to be β = 0.0003 for CIFAR datasets and β = 0.0001 for SVHN. |