BulletTrain: Accelerating Robust Neural Network Training via Boundary Example Mining

Authors: Weizhe Hua, Yichi Zhang, Chuan Guo, Zhiru Zhang, G. Edward Suh

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

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our technique to several existing robust training algorithms and achieve a 2.2 speed-up for TRADES and MART on CIFAR-10 and a 1.7 speed-up for Aug Mix on CIFAR-10-C and CIFAR-100-C without any reduction in clean and robust accuracy. 4 Experiments
Researcher Affiliation Collaboration Weizhe Hua1, Yichi Zhang1, Chuan Guo2, Zhiru Zhang1, G. Edward Suh1,2 1Cornell University, 2Facebook AI Research
Pseudocode Yes Algorithm 1: Standard robust DNN training algorithm. Algorithm 2: Bullet Train.
Open Source Code No The paper does not contain any explicit statements about releasing source code, nor does it provide a link to a code repository.
Open Datasets Yes We apply our technique to several existing robust training algorithms and achieve a 2.2 speed-up for TRADES and MART on CIFAR-10 and a 1.7 speed-up for Aug Mix on CIFAR-10-C and CIFAR-100-C without any reduction in clean and robust accuracy. MNIST (Le Cun et al., 2010) CIFAR-10 (Krizhevsky, 2009).
Dataset Splits No The paper mentions 'training set' and 'corrupted test samples' for evaluation, implying train/test splits, but it does not specify exact percentages, sample counts for training, validation, and test splits, nor does it explicitly mention a validation set.
Hardware Specification No The paper states 'using a single NVIDIA GPU' but does not provide specific model numbers or detailed specifications of the hardware used for experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes When Bullet Train is applied, we set NB = 10 with α = 0.007, NR ∈ [0, 2] with α = 1.7ϵ/NR, NO = 0, and γ = 0.8. Figure 6 shows that the accuracy and theoretical speedup of Bullet Train under different γ ∈ [0.65, 0.9].