AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

On the Importance of Gradients for Detecting Distributional Shifts in the Wild

Rui Huang, Andrew Geng, Yixuan Li

NeurIPS 2021 · Label-efficient and robust learning › OOD and anomaly detection · approach: Spurious-aware OOD detection · 38 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 gradient norm of KL(uniform||softmax) instead of Energy scoreEnergy · ImageNet-1k OOD (ResNetv2-101)FPR95: 71.03 -> 54.70Improved
use GradNorm instead of Energy score on DenseNet-121Energy · ImageNet-1k OOD (DenseNet-121)FPR95: 51.29 -> 41.00Improved
use gradients from last FC layer only instead of all parameters (ablation)GradNorm · ImageNet-1k OOD avgFPR95: 69.35 -> 54.70Improved
use gradients from last layer instead of Block 1 (ablation)GradNorm · ImageNet-1k OOD avgFPR95: 73.52 -> 54.70Improved
use uniform target (all labels) instead of one-hot target for backpropagation (ablation)GradNorm · SUNFPR95: reduces FPR95 by 48.35%Improved
use L1 norm instead of higher-order Lp norms (ablation)GradNorm · ImageNet-1k OODFPR95: up to 22.31% FPR95 over L2Improved
change temperature from T=1 (ablation)GradNorm · ImageNet-1k OODFPR95Hurt
use gradient norm instead of KL divergence directly from output space (ablation)KL divergence score · ImageNet-1k OODFPR95Improved
combine feature term U and output term V (U*V) instead of either alone (ablation)GradNorm · ImageNet-1k OOD avgFPR95: 71.00 -> 54.70 (U alone), 73.18 -> 54.70 (V alone)Improved
increase backbone capacity beyond ResNetv2-101x1 (ablation)GradNorm · ImageNet-1k OOD avgFPR95: 55.84 -> 66.20 (101x3)Hurt
use GradNorm instead of MahalanobisMahalanobis · ImageNet-1k OOD (ResNetv2-101)FPR95: 81.69 -> 54.70Improved

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