AI Paper Atlas

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

Hessian-based Analysis of Large Batch Training and Robustness to Adversaries

Zhewei Yao, Amir Gholami, Qi Lei, Kurt Keutzer, Michael W. Mahoney

NeurIPS 2018 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training recipes · 95 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
Increase C1 training batch size from 16 to 2048C1 · CIFAR-10 testaccuracy: 77.68 -> 77.31Mixed
Increase C1 training batch size from 16 to 2048C1 · CIFAR-10 test, ε=0.02adversarial accuracy: 30.38 -> 3.77Hurt
Increase CR training batch size from 64 to 256CR · CIFAR-100 testaccuracy: 70.81 -> 68.6Hurt
Increase C1 training batch size from 256 to 16384C1 · SVHN testaccuracy: 95.70 -> 94.86Hurt
Use FGSM adversarial training instead of ordinary trainingMNIST model · MNIST adversarial samples, mean of attacksaccuracy: 53.44 -> 87.59Improved
Use FGSM adversarial training instead of ordinary trainingCIFAR-10 model · CIFAR-10 adversarial samples, mean of attacksaccuracy: 16.93 -> 62.51Improved
Use FGSM adversarial training instead of ordinary trainingMNIST model · MNIST clean testaccuracy: 99.32 -> 99.49No clear effect
Replace ReLU with Softplus (β=20) (ablation)C1 · CIFAR-10 test, batch 128 vs 2048accuracy: 78.79 -> 76.27Hurt
Replace ReLU with ELU (α=1) (ablation)C1 · CIFAR-10 test, batch 128 vs 2048accuracy: 78.94 -> 76.64Hurt
Replace ReLU with Softplus (β=20) (ablation)C1 · CIFAR-10 adversarial test, ε=0.02, batch 128 vs 2048adversarial accuracy: 17.79 -> 2.29Hurt
Replace ReLU with ELU (α=1) (ablation)C1 · CIFAR-10 adversarial test, ε=0.02, batch 128 vs 2048adversarial accuracy: 15.26 -> 3.99Hurt

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