High-dimensional and permutation invariant anomaly detection

Mikuni, Vinicius; Nachman, Benjamin

doi:10.21468/SciPostPhys.16.3.062

SciPost Physics

High-dimensional and permutation invariant anomaly detection

Vinicius Mikuni, Benjamin Nachman

SciPost Phys. 16, 062 (2024) · published 29 February 2024

doi: 10.21468/SciPostPhys.16.3.062
pdf
Submissions/Reports

Abstract

Methods for anomaly detection of new physics processes are often limited to low-dimensional spaces due to the difficulty of learning high-dimensional probability densities. Particularly at the constituent level, incorporating desirable properties such as permutation invariance and variable-length inputs becomes difficult within popular density estimation methods. In this work, we introduce a permutation-invariant density estimator for particle physics data based on diffusion models, specifically designed to handle variable-length inputs. We demonstrate the efficacy of our methodology by utilizing the learned density as a permutation-invariant anomaly detection score, effectively identifying jets with low likelihood under the background-only hypothesis. To validate our density estimation method, we investigate the ratio of learned densities and compare to those obtained by a supervised classification algorithm.

TY  - JOUR
PB  - SciPost Foundation
DO  - 10.21468/SciPostPhys.16.3.062
TI  - High-dimensional and permutation invariant anomaly detection
PY  - 2024/02/29
UR  - https://scipost.org/SciPostPhys.16.3.062
JF  - SciPost Physics
JA  - SciPost Phys.
VL  - 16
IS  - 3
SP  - 062
A1  - Mikuni, Vinicius
AU  - Nachman, Benjamin
AB  - Methods for anomaly detection of new physics processes are often limited to low-dimensional spaces due to the difficulty of learning high-dimensional probability densities. Particularly at the constituent level, incorporating desirable properties such as permutation invariance and variable-length inputs becomes difficult within popular density estimation methods. In this work, we introduce a permutation-invariant density estimator for particle physics data based on diffusion models, specifically designed to handle variable-length inputs. We demonstrate the efficacy of our methodology by utilizing the learned density as a permutation-invariant anomaly detection score, effectively identifying jets with low likelihood under the background-only hypothesis. To validate our density estimation method, we investigate the ratio of learned densities and compare to those obtained by a supervised classification algorithm.
ER  -

@Article{10.21468/SciPostPhys.16.3.062,
	title={{High-dimensional and permutation invariant anomaly detection}},
	author={Vinicius Mikuni and Benjamin Nachman},
	journal={SciPost Phys.},
	volume={16},
	pages={062},
	year={2024},
	publisher={SciPost},
	doi={10.21468/SciPostPhys.16.3.062},
	url={https://scipost.org/10.21468/SciPostPhys.16.3.062},
}

Cited by 6

Authors / Affiliations: mappings to Contributors and Organizations

See all Organizations.

¹ Vinicius Mikuni,
¹ ² Benjamin Nachman

Funders for the research work leading to this publication