AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Red Teaming Deep Neural Networks with Feature Synthesis Tools

Stephen Casper, Tong Bu, Yuxiao Li, Jiawei Li, Kevin Zhang, Kaivalya Hariharan, Dylan Hadfield-Menell

NeurIPS 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Self-explaining and visual explanation networks · 1 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
Use feature synthesis instead of feature attribution/saliency for trojan discoveryInterpretability tools · 12-trojan benchmarkhuman trigger identificationImproved
Use robust feature-level adversaries or SNAFUE rather than other individual synthesis toolsFeature synthesis tools · 12-trojan benchmarkhuman trigger identificationImproved
Combine visualizations from multiple tools rather than relying on one toolInterpretability toolbox · 12-trojan benchmarkhuman trigger identificationImproved
Visualize target neurons instead of multiple associated inner neurons (ablation)Feature visualization · Trojan trigger identificationhuman trigger identificationImproved
Use occlusion rather than an edge detector baselineFeature attribution/saliency · Patch trojans on ResNet50 and VGG19pixel-wise Pearson correlationImproved
Use Fourier-space feature visualization or TABORFeature synthesis tools · 12-trojan benchmarkhuman trigger identificationNo clear effect
Use feature synthesis methods to rediscover style-transfer trojansFeature synthesis tools · Style trojanshuman trigger identificationHurt
Use human evaluators instead of CLIP to judge visualizationsInterpretability tool evaluation · 12-trojan benchmarkcorrect trigger identificationImproved
Use SNAFUE for copy/paste attacks between dissimilar source and target classesSNAFUE · Clothing-to-traffic ImageNet attacksnatural patch attack success: Only one out of the 1,000 total natural patches ... succeeded for at least 50% of source class images; 54% ... were succMixed
Use SNAFUE to recover an implanted trigger included among candidate patchesSNAFUE · Four patch-trojan casestrigger recovered in top 10: 2 of the 4 casesMixed

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