AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Window-Based Distribution Shift Detection for Deep Neural Networks

Guy Bar-Shalom, Yonatan Geifman, Ran El-Yaniv

NeurIPS 2023 · Learning theory and algorithms › Prediction and decision losses · approach: Probabilistic calibration · 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 coverage-based window detector instead of KS-Softmaxwindow shift detector · ResNet50, window 10AUROC: 61 -> 78Improved
use coverage-based detector instead of KS-Softmax (ViT-T, small window)window shift detector · ViT-T, window 10AUROC: 62 -> 89Improved
use coverage-based detector instead of KS-Softmax at window 1000window shift detector · ResNet50, window 1000AUROC: 100 -> 100No clear effect
use coverage-based detector instead of KS-Softmax on MobileNetV3-S at window 100window shift detector · MobileNetV3-S, window 100Detection Error: 10 -> 11Hurt
use single-instance baselines (SR, Entropy) for population-level detectionsingle-instance t-test · ResNet50, window 1000AUROC: 88 -> 100Hurt
use coverage detector instead of KS-Softmax under FGSM shiftwindow shift detector · ResNet50 FGSM eps=1e-4, window 50AUROC: 64 -> 88Improved
use coverage detector instead of MMD-Embeddings on ImageNet-O, window 10window shift detector · ResNet50 ImageNet-O, window 10AUROC: 96 -> 61Hurt
use coverage detector instead of KS-Softmax under Zoom out 90% shiftwindow shift detector · ResNet50 Zoom 90%, window 50AUROC: 61 -> 81Improved
use coverage-based detection instead of KS-Softmax for run-timewindow shift detector · detection-training set 1,000,000detection time (s): 95.3 -> 4.66 \cdot 10^{-4}Improved
use a single target coverage C_target=1 instead of multiple coverages (ablation)coverage detector · ResNet50average AUROCHurt
increase number of target coverages C_target (1 vs 10 vs 100) (ablation)coverage detector · ResNet50average AUROCImproved
vary confidence level delta over 0.1, 0.01, 0.001, 0.0001 (ablation)coverage detector · ResNet50average AUROCNo clear effect

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