Improving Robustness using Generated Data

Authors: Sven Gowal, Sylvestre-Alvise Rebuffi, Olivia Wiles, Florian Stimberg, Dan Andrei Calian, Timothy A Mann

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on CIFAR-10, CIFAR-100, SVHN and TINYIMAGENET against ℓ and ℓ2 norm-bounded perturbations of size ϵ = 8/255 and ϵ = 128/255, respectively. We show large absolute improvements in robust accuracy compared to previous state-of-the-art methods.
Researcher Affiliation Industry Sven Gowal*, Sylvestre-Alvise Rebuffi*, Olivia Wiles, Florian Stimberg, Dan Calian and Timothy Mann Deep Mind, London {sgowal,sylvestre}@deepmind.com
Pseudocode No The information provided does not contain a structured pseudocode or algorithm block. The overall approach is described verbally and summarized in Figure 2, but not as a formal algorithmic listing.
Open Source Code Yes The training and evaluation code is available at https://github.com/deepmind/deepmind-research/tree/master/ adversarial_robustness.
Open Datasets Yes We use the CIFAR-10 and CIFAR-100 datasets [42], as well as SVHN [53] and TINYIMAGENET [26].
Dataset Splits Yes We perform hyper-parameter tuning on a held-out validation set.
Hardware Specification Yes All experiments are run on NVIDIA A100 GPUs.
Software Dependencies No The information provided mentions software used ('The models are implemented in Jax [6] and Haiku [35]') but does not specify version numbers for these or any other key software components, which is required for reproducible description of ancillary software.
Experiment Setup Yes We use stochastic weight averaging [38] with a decay rate of 0.995. For adversarial training, we use TRADES [82] with 10 Projected Gradient Descent (PGD) steps. We train for 400 CIFAR-10-equivalent epochs with a batch size of 1024 (i.e., 19K steps).