AI Paper Atlas

Research map / Time series / Anomaly detection

EVEREST: A Transformer for Probabilistic Rare-Event Anomaly Detection with Evidential and Tail-Aware Uncertainty

Antanas Žilinskas, Robert Noel Shorten, Jakub Marecek

ICLR 2026 · Time series › Anomaly detection · approach: Multivariate anomaly detection

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
add attention bottleneck instead of mean pooling (ablation)EVEREST · M5-72hTSS: 0.319 -> 0.746; +0.427 TSSImproved
add EVT head (ablation)EVEREST · M5-72hTSS: 0.461 -> 0.746; +0.285 TSSImproved
add evidential NIG head (ablation)EVEREST · M5-72hTSS: 0.682 -> 0.746; +0.064 TSSImproved
add precursor auxiliary head (ablation)EVEREST · M5-72hTSS: -0.650 TSSImproved
use focal loss instead of cross-entropy (ablation)EVEREST · M5-72hTSS: -0.537 TSSImproved
use mixed precision (AMP) instead of FP32 (ablation)EVEREST · M5-72hTSS: 0.000 -> 0.746Improved
add composite schedule with gamma annealing and joint training (ablation)EVEREST · M5-72hTSS: +0.045 TSSImproved
add evidential head (calibration effect) (ablation)EVEREST · M5-72hECE: 0.0111 -> 0.0110No clear effect
add EVT head (calibration effect) (ablation)EVEREST · M5-72hECE: 0.0336 -> 0.0110Improved
vary EVT exceedance quantile u in {0.85, 0.90, 0.95} (ablation)EVEREST · M5-72hTSS: TSS 0.903 to 0.932; ECE 0.0119 to 0.0164No clear effect
vary evidential and EVT loss weights over 5x5 grid (0.5 to 1.5 multipliers) (ablation)EVEREST · M5-72hTSS and ECENo clear effect
apply EVEREST to SKAB valve anomalies without task-specific tuningEVEREST · SKABF1 (%): 98.2 ± 1.7 vs 91–96Improved

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