AI Paper Atlas

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

SmoothMix: Training Confidence-calibrated Smoothed Classifiers for Certified Robustness

Jongheon Jeong, Sejun Park, Minkyu Kim, Heung-Chang Lee, Doguk Kim, Jinwoo Shin

NeurIPS 2021 · Adversarial robustness and security › Certified and empirical robustness · approach: Randomized smoothing certification · 13 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
Add SmoothMix self-mixup loss at η=5.0 instead of η=1.0 (ablation)SmoothMix · MNIST test, σ=1.00ACR: 1.788 -> 1.820Improved
Use a one-step adversary as the training input (ablation)SmoothMix · MNIST test, σ=1.00ACR: 1.788 -> 1.816Improved
Use a one-step adversary as the training input (ablation)SmoothMix · CIFAR-10 test, σ=0.50ACR: 0.720 -> 0.737Improved
Use a one-step adversary as the training input (ablation)SmoothMix · CIFAR-10 test, σ=0.25ACR: 0.553 -> 0.548Hurt
Increase the relative weight η of the SmoothMix loss (ablation)SmoothMix · MNIST, σ=1.0certified robustnessImproved
Keep the product α·T constant while varying step size and number of steps (ablation)SmoothMix · MNIST, σ=1.0ACRNo clear effect
Add a hard ℓ2 radius restriction to adversarial-example search (ablation)SmoothMix · MNIST, σ=1.0certified robustnessHurt
Reduce the number of noise samples used to estimate the smoothed prediction to m=1 (ablation)SmoothMix · MNIST, σ=1.0ACRNo clear effect
Train with SmoothMix rather than Gaussian augmentation aloneSmoothed classifier · MNIST, σ=1.0ACRImproved
Train with SmoothMix rather than ConsistencySmoothed classifier · CIFAR-10, σ=0.50ACR: 0.720 -> 0.737Improved

Later papers that compared with this method

5 later accepted papers report SmoothMix as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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