|
Activity Number:
|
31
- Statistical Inference of Causality and Structure
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 7, 2022 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
IMS
|
|
Abstract #323095
|
|
|
Title:
|
A One-Sided Multinomial Hypothesis Test for Unsupervised Anomaly Detection
|
|
Author(s):
|
Danielle Gewurz* and Bill Roberts and Lun Li and Morgan DeHart
|
|
Companies:
|
Deloitte Consulting and Deloitte Consulting and Deloitte Consulting and Deloitte Consulting
|
|
Keywords:
|
anomaly detection;
one-sided test;
multinomial;
water filling;
unsupervised;
likelihood ratio test
|
|
Abstract:
|
A multivariate analogue of a one-sided test is presented for multinomial distributions. The test involves a constrained optimization that is formulated using Karush-Kuhn-Tucker conditions. An explicit solution to the constrained optimization is derived by adapting the water filling algorithm encountered in information theory. Feasibility of the resulting test is demonstrated using standardized anomaly detection tasks from the literature.
|
Authors who are presenting talks have a * after their name.
