AI Paper Atlas

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

A Theoretical View of Linear Backpropagation and its Convergence

Ziang Li, Yiwen Guo, Haodi Liu, Changshui Zhang

TPAMI 2024 · Adversarial robustness and security › Black-box and transfer attacks · approach: Transfer attacks on face recognition · 5 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 LinBP instead of BP for gradientsPGD white-box attack · CIFAR-10 ResNet-50, K=10, eps=8/255attack success rate: 66.41% -> 85.83%Improved
use LinBP instead of BP in PGD with 100 stepsPGD white-box attack · CIFAR-10 non-robust models, K=100attack success rate: 99.73% -> 100.00%No clear effect
use LinBP instead of BPPGD attack on robust WRN-28-10 · CIFAR-10 K=10 eps=8/255attack success rate: 33.14% -> 33.48%Improved
use LinBP instead of BPAuto-PGD attack · CIFAR-10 ResNet-50 eps=8/255attack success rate: 95.77% -> 97.60%Improved
use LinBP so Auto-PGD halves step size soonerAuto-PGD on ResNet-50 · CIFAR-10 eps=8/255iterations before first step halving: 22.95 vs 25.85Improved
use LinBP with Swish/SiLU activationsPGD on ResNet-50 (SiLU) · CIFAR-10 K=10 eps=16/255attack success rate: 82.65% -> 87.97%Improved
use LinBP instead of BP gradients with l2 gradient normalization in simulated model trainingone-layer teacher-student · synthetic trainingl1 distance to teacher: similar training loss and approximationNo clear effect
use LinBP gradients for trainingMLP and LeNet-5 SGD · MNISTtraining loss and accuracy: lower loss, higher accuracy earlyImproved
use LinBP gradients for trainingResNet-50, DenseNet-161, MobileNetV2 · CIFAR-10final accuracy: comparable to BP at the endNo clear effect
scale learning rate 10x with LinBPLeNet-5 SGD · MNISTtraining convergence: failed to converge with LinBP while BP convergedHurt
use LinBP gradients for training with normalized gradientsDNN training · CIFAR-10/MNISTtraining performance: LinBP no better than BPNo clear effect
generate adversarial examples with LinBP during adversarial training (second strategy)adversarial training · CIFAR-10 robustness to BP-generated examplestest robust accuracy: larger generalization gap, degraded robustnessHurt

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