When does Contrastive Learning Preserve Adversarial Robustness from Pretraining to Finetuning?

Authors: Lijie Fan, Sijia Liu, Pin-Yu Chen, Gaoyuan Zhang, Chuang Gan

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

Reproducibility Variable Result LLM Response
Research Type Experimental With a thorough experimental study, we demonstrate that ADVCL outperforms the state-of-the-art self-supervised robust learning methods across multiple datasets (CIFAR-10, CIFAR-100 and STL-10) and finetuning schemes (linear evaluation and full model finetuning).
Researcher Affiliation Collaboration Lijie Fan1, Sijia Liu2, Pin-Yu Chen3, Gaoyuan Zhang3, Chuang Gan3 1 Massachusetts Institute of Technology, 2 Michigan State University, 3 MIT-IBM Watson AI Lab, IBM Research
Pseudocode No The paper provides a pipeline diagram (Figure 2) but no structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/Lijie Fan/Adv CL.
Open Datasets Yes With a thorough experimental study, we demonstrate that ADVCL outperforms the state-of-the-art self-supervised robust learning methods across multiple datasets (CIFAR-10, CIFAR-100 and STL-10) and finetuning schemes (linear evaluation and full model finetuning).
Dataset Splits No The paper mentions training and testing, and discusses finetuning schemes, but does not explicitly describe a validation dataset split or how it was used.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU models, CPU types, memory amounts) used for running its experiments. It only mentions using ResNet-18 as an encoder architecture.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., Python 3.8, PyTorch 1.9, CUDA 11.1) needed to replicate the experiment.
Experiment Setup Yes Unless specified otherwise, we use 5-step 1 projected gradient descent (PGD) with = 8/255 to generate perturbations during pretraining, and use Auto-Attack and 20-step 1 PGD with = 8/255 to generate perturbations in computing AA and RA at test time.