AI Paper Atlas

Research map / Adversarial robustness and security / Black-box and transfer attacks

Making Adversarial Examples More Transferable and Indistinguishable

Junhua Zou, Yexin Duan, Boyu Li, Wu Zhang, Yu Pan, Zhisong Pan

AAAI 2022 · Adversarial robustness and security › Black-box and transfer attacks · approach: Transfer attacks on face recognition · 29 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
Replace the sign function with tanh (ablation)TI-DIM · Inc-v3 attack; six classic defensesmean perturbation: 10.46 -> 9.14Improved
Replace the sign function with tanh (ablation)TI-DIM · Inc-v3 attack; six classic defensesmean success rate (%): 82.0 -> 82.4Improved
Use Adam instead of the baseline gradient update (ablation)TI-DIM · Inc-v3 attack; six classic defensesmean success rate (%): 82.0 -> 83.6Improved
Use Adam instead of the baseline gradient update (ablation)TI-DIM · Inc-v3 attack; six classic defensesmean perturbation: 10.46 -> 9.20Improved
Use tanh together with Adam (ablation)TI-DIM · Inc-v3 attack; six classic defensesmean perturbation: 10.46 -> 7.86Improved
Use tanh together with Adam (ablation)TI-DIM · Inc-v3 attack; six classic defensesmean success rate (%): 82.0 -> 83.1Improved
Use smaller Gaussian kernels (ablation)TI-DIM + tanh + Adam · Inc-v3 attack; six classic defensesmean success rate (%): 83.1 -> 86.5Improved
Use increasing step sizes (ablation)TI-DIM + tanh + Adam + smaller kernels · Inc-v3 attack; six classic defensesmean success rate (%): 86.5 -> 88.0Improved
Use increasing step sizes (ablation)TI-DIM + tanh + Adam + smaller kernels · Inc-v3 attack; six classic defensesmean perturbation: 7.82 -> 8.11Hurt
Replace TI-DIM with TI-DI-AITMEnsemble attacks · Six classic defensesmean attack success rate (%): 82.0 -> 88.0Improved
Replace NI-TI-DIM with NI-TI-DI-AITMEnsemble attacks · Six classic defensesmean attack success rate (%): 83.7 -> 89.3Improved
Replace NI-TI-DIM with NI-TI-DI-AITMEnsemble attacks · Three advanced defensesmean attack success rate (%): 75.1 -> 82.7Improved

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