AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Adversarial Attack for Asynchronous Event-Based Data

Wooju Lee, Hyun Myung

AAAI 2022 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial perturbations beyond images · 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
add generation of adversarial events to event-time shifting (ablation)Event attack · N-Caltech101, untargeted, most tested representationsattack success rate (%): 95.50 -> 95.87; 96.13 -> 100.0; 95.96 -> 99.0; 95.14 -> 96.12; 94.16 -> 96.03Improved
add generation of adversarial events to event-time shifting (ablation)Event attack · N-Caltech101, random targeted, EST (10), EST (5), Voxel grid (10), MLP and EST (10), Trilinearattack success rate (%): 41.64 -> 69.50; 63.87 -> 70.74; 52.62 -> 80.23; 41.42 -> 69.78Improved
add generation of adversarial events to shifting (ablation)Event attack · N-Caltech101, random targeted, Two-channel (1)attack success rate (%): 69.87 -> 69.48Hurt
add generation of adversarial events to shifting (ablation)Event attack · N-Caltech101, random targeted, Two-channel (1), Trilinearattack success rate (%): 67.84 -> 67.42Hurt
use adversarial training on original and adversarial eventsEST (5) ResNet-34 · N-Caltech101, shifting attacktop-1 accuracy (%): 2.99 -> 52.79Improved
use adversarial training on original and adversarial eventsEST (5) ResNet-34 · N-Caltech101, shifting and generating attacktop-1 accuracy (%): 0 -> 22.86Improved
use adversarial training rather than original-only trainingEST (5) ResNet-34 · N-Caltech101, cleantop-1 accuracy (%): 84.99 -> 80.67Hurt
increase the shifting perturbation size (ablation)EST (10) · N-Caltech101, untargeted attackattack success rate (%): 95.11; 95.50; 93.40; 94.96; 90.66Mixed
increase the shifting perturbation size (ablation)EST (1) · N-Caltech101, untargeted attackattack success rate (%): 62.61; 86.01; 93.63; 95.96; 98.05Improved
reduce event-data frequency by half (ablation)EST (10) shifting attack · N-Caltech101, untargeted attack, epsilon 0.05attack success rate (%): 95.11 -> 75.5Hurt
use adversarial events from EST (10) or Voxel grid (10) against EST (1)Cross-representation attack · N-Caltech101, untargeted transferattack success rate (%): 0.72; 0.86Hurt
use adversarial events from EST (1) against other representationsCross-representation attack · N-Caltech101, untargeted transferattack success rate (%): 27.51; 46.85Improved

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