Outlier Synthesis via Hamiltonian Monte Carlo for Out-of-Distribution Detection

Hengzhuang Li, Teng Zhang · 2025

Read Outlier Synthesis via Hamiltonian Monte Carlo for Out-of-Distribution Detection by Hengzhuang Li, Teng Zhang (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-2501-16718.txt.

Description

Hengzhuang Li and Teng Zhang propose, in a 2025 paper, a way to make fake odd cases that teach a model to spot real ones. A model that must flag data unlike its training set works best when it has seen odd cases in training. Good real odd cases are hard to gather at scale. Past fake-case methods made poor cases or cost much to run. Their frame, HamOS, treats fake-case making as a walk through trait space, led by a Markov chain.

The walk starts from in-set data alone and roams wide, so the fake cases span many odd scenes. A sampling rule with a near-full accept rate keeps the walk cheap. On standard and large bench sets, the method beat top rivals on both hit rate and speed. The team casts it as a lean path to safer models.

Catalog record

CREATORHengzhuang Li, Teng Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE142 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-16718

Browse the full library · About the Tower of Babel