Scalable, Explainable and Provably Robust Anomaly Detection with One-Step Flow Matching

Zhong Li, Qi Huang, Yuxuan Zhu, Lincen Yang, Mohammad Mohammadi Amiri, Niki van Stein, Matthijs van Leeuwen · 2025

Read Scalable, Explainable and Provably Robust Anomaly Detection with One-Step Flow Matching by Zhong Li, Qi Huang, Yuxuan Zhu, Lincen Yang, Mohammad Mohammadi Amiri, Niki van Stein, Matthijs van Leeuwen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-18328.txt.

Description

Zhong Li and his six co-authors present a method for spotting anomalies, or unusual rows, in tables of data. Their method, Time-Conditioned Contraction Matching, borrows an idea from flow matching, a recent way of learning how one spread of data can move toward another. Rather than solving heavy equations at test time, the new method learns to point each input back toward a fixed center.

A single quick pass then scores each row by how far its movement departs from the expected pattern, and the score can be traced back to individual columns. The authors also prove the score changes only in a bounded way when inputs shift slightly, which supports trust in small cases. On the ADBench test set, the method balanced accuracy and speed well, with the strongest gains on large and wide data sets.

Catalog record

CREATORZhong Li, Qi Huang, Yuxuan Zhu, Lincen Yang, Mohammad Mohammadi Amiri, Niki van Stein, Matthijs van Leeuwen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE275 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-18328

Browse the full library · About the Tower of Babel