mlsanityData to 7 Oct 2026

How Many Categories Are Enough? Distribution-Free Certification Limits for Few-Shot Anomaly Thresholds

arXiv 2610.00236 ↗First posted 23 Sep 2026cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Few-shot anomaly detectors are judged by ranking metrics, yet deployment requires an alarm threshold with a controlled false-alarm rate (FAR). We ask how much normal evidence, in images or category units, is needed to certify such a threshold for an unseen category. Using a frozen DINOv2 principal component analysis (PCA) residual ranker on 15 MVTec and 12 VisA categories under four corruption types, we show that target-only leave-one-image-out (LOIO) calibration is resolution-limited and shift-fragile: rank values cannot fall below 1/(k+1), and at the attainable level α=0.20, empirical FAR reaches 0.341 on Gaussian-corrupted MVTec at k=4, 1.7 times the nominal level. A category-count feasibility calculus is then derived: even with all-zero category losses and no multiplicity charged, any deterministic, uniformly valid, distribution-free 95% upper confidence bound (UCB) requires at least 14, 29, and 59 independent and identically distributed (iid) category draws at α=0.20, 0.10, and 0.05; these counts are necessary but not sufficient. The Cross-category Reliability Estimation with Source Support (CRESS) protocol splits source categories into disjoint reference, proposal, and certification roles. With only three or four certification categories, all 960 frozen configurations return the fail-closed threshold τ^star=0, and the smallest category-level UCB is 0.950. Image-unit analyses of the same archive select positive thresholds in 36.7% to 60.3% of target cells; these bounds hold for the selected source mixture, not for the marginal risk of a new-category draw. The contribution is a quantitative feasibility boundary and an estimand-aware protocol specifying when source evidence can, and cannot, support a transferable reliability claim.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 20 of its 34 references that are in the map. First posted on arXiv 23 Sep 2026.

Builds on

20 of 34 references.

References · 20

  1. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  2. 2017On Calibration of Modern Neural Networkscited by 4,860
  3. 2021A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantificationcited by 777
  4. 2020PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localizationcited by 591
  5. 2021CutPaste: Self-Supervised Learning for Anomaly Detection and Localizationcited by 529
  6. 2021DRAEM -- A discriminatively trained reconstruction embedding for surface anomaly detectioncited by 510
  7. 2022Conformal Risk Controlcited by 290
  8. 2021Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Controlcited by 256
  9. 2019Predictive inference with the jackknife+cited by 206
  10. 2021Testing for Outliers with Conformal p-valuescited by 133
  11. 2022Registration based Few-Shot Anomaly Detectioncited by 133
  12. 2024Toward Generalist Anomaly Detection via In-context Residual Learning with Few-shot Sample Promptscited by 88
8 more
  1. 2023Pushing the Limits of Fewshot Anomaly Detection in Industry Vision: Graphcorecited by 75
  2. 2024AnomalyDINO: Boosting Patch-based Few-shot Anomaly Detection with DINOv2cited by 53
  3. 2022PAC-Wrap: Semi-Supervised PAC Anomaly Detectioncited by 10
  4. 2026SubspaceAD: Training-Free Few-Shot Anomaly Detection via Subspace Modelingcited by 7
  5. 2024Leave-One-Out-, Bootstrap- and Cross-Conformal Anomaly Detectorscited by 4
  6. 2026Between Resolution Collapse and Variance Inflation: Weighted Conformal Anomaly Detection in Low-Data Regimescited by 4
  7. 2026Conformal Anomaly Detection in Python: Moving Beyond Heuristic Thresholds with nonconformcited by 1
  8. 2026HeAD-CP: Heterophily-Aware Diffused Conformal Prediction Sets for Graph Neural Networkscited by 1

Built on by

Not cited yet.