AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

BulletTrain: Accelerating Robust Neural Network Training via Boundary Example Mining

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

NeurIPS 2021 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training recipes · 6 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply BulletTrain to AugMix (JSD loss on 35% of clean examples)AugMix · CIFAR-10-Crobust accuracy: 88.8% -> 88.8%No clear effect
apply BulletTrain to AugMixAugMix · CIFAR-100-Crobust accuracy: 64.0% -> 63.6%No clear effect
remove JSD loss for all examples instead of selective BulletTrain (ablation)AugMix · CIFAR-10-Crobust accuracy: 88.8% -> 86.9%Hurt
remove JSD loss for all examples (ablation)AugMix · CIFAR-100-Crobust accuracy: 64.0% -> 60.2%Hurt
apply BulletTrain N_R=2 to TRADES lambda=1/6TRADES · CIFAR-10 PGD-20robust accuracy: 56.61% -> 56.35%No clear effect
apply BulletTrain N_R=2 to TRADES lambda=1/6TRADES · CIFAR-10clean accuracy: 84.92% -> 85.89%Improved
reduce N_R from 2 to 0 in BulletTrain instead of lowering lambda (ablation)TRADES · CIFAR-10 PGD-20robust accuracy: 56.35% -> 52.10%Hurt
apply BulletTrain N_R=2 to MARTMART · CIFAR-10 PGD-20robust accuracy: 57.39% -> 58.74%Improved
apply BulletTrain N_R=1 to MARTMART · CIFAR-10 PGD-20robust accuracy: 57.39% -> 58.11%Improved
replace multi-step PGD with YOPO-2-5 in TRADESTRADES · CIFAR-10 PGD-20robust accuracy: 45.28% vs 49.14%Hurt
use single-step FAST training with TRADESTRADES · CIFAR-10 PGD-20robust accuracy: 56.61% -> 4.48%Hurt
use single-step FAST training with MARTMART · CIFAR-10 PGD-20robust accuracy: 57.39% -> 0.20%Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy