AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Attribution-Guided Model Rectification of Unreliable Neural Network Behaviors

Peiyu Yang, Naveed Akhtar, Jiantong Jiang, Ajmal Mian

CVPR 2026 · Architectures and efficiency › RNNs and normalization · approach: Layer-wise analysis of language models · 3 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 dynamic layer localization instead of final layer only (ablation)rank-one rectifying (n=1) · CIFAR-10 TrojanASR (%): 2.57 -> 1.34Improved
use dynamic layer localization instead of final layer only (ablation)rank-one rectifying (n=1) · ImageNet TrojanASR (%): 3.01 -> 1.61Improved
use dynamic layer localization instead of final layer only (ablation)rank-one rectifying (n=1) · ImageNet Trojanoverall accuracy (%): 67.87 -> 66.77Hurt
use dynamic layer localization instead of final layer only (ablation)rank-one rectifying (n=1) · CIFAR-10 spurious correlationspurious-clean gap: 2.23 -> 1.86Improved
use dynamic layer localization instead of final layer only (ablation)rank-one rectifying (n=1) · BlockMNISTfeature leakage: 2.655 -> 2.417Improved
use dynamic layer localization instead of final layer only (ablation)rank-one rectifying · ISICspurious-clean gap: 4.50 -> 1.50Improved
rank-one rectifying instead of fine-tuningTrojaned CIFAR-10 model · CIFAR-10 TrojanASR (%): 13.22 -> 0.26Improved
rank-one rectifying instead of P-ClArCTrojaned CIFAR-10 model · CIFAR-10 Trojanoverall accuracy (%): 89.97 -> 93.61Improved
increase cleansed samples from 1 to 20 (ablation)Dyn. rectifying · CIFAR-10 TrojanASR (%): 1.34 -> 0.26Improved
increase cleansed samples from 1 to 20 (ablation)Dyn. rectifying · ImageNet Trojanoverall accuracy (%): 66.77 -> 68.84Improved
fine-tune instead of rectifybenign model · CIFAR-10 spurious correlationspurious-clean gap: 5.58 -> 10.65Hurt
add IG regularizationbenign BlockMNIST model · BlockMNISTaccuracy: 99.17 -> 94.14Hurt

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