Anomaly Detection and Approximate Matching via Entropy Divergences

Authors: Russell Leidich

The Jensen-Shannon divergence (JSD) quantifies the “information distance” between a pair of probability distributions. (A more generalized version, which is beyond the scope of this paper, is given in [1]. It extends this divergence to arbitrarily many such distributions. Related divergences are presented in [2], which is an excellent summary of existing work.)

A couple of novel applications for this divergence are presented herein, both of which involving sets of whole numbers constrained by some nonzero maximum value. (We’re primarily concerned with discrete applications of the JSD, although it’s defined for analog variables.) The first of these, which we can call the “Jensen-Shannon divergence transform” (JSDT), involves a sliding “sweep window” whose JSD with respect to some fixed “needle” is evaluated at each step as said window moves from left to right across a superset called a “haystack”.

The second such application, which we can call the “Jensen-Shannon exodivergence transform” (JSET), measures the JSD between a sweep window and an “exosweep”, that is, the haystack minus said window, at all possible locations of the latter. The JSET turns out to be exceptionally good at detecting anomalous contiguous subsets of a larger set of whole numbers.

We then investigate and attempt to improve upon the shortcomings of the JSD and the related Kullback-Leibler divergence (KLD).

Comments: 20 Pages.

Download: PDF

Submission history

[v1] 2017-10-22 23:43:04
[v2] 2017-12-01 10:20:35

Unique-IP document downloads: 105 times is a pre-print repository rather than a journal. Articles hosted may not yet have been verified by peer-review and should be treated as preliminary. In particular, anything that appears to include financial or legal advice or proposed medical treatments should be treated with due caution. will not be responsible for any consequences of actions that result from any form of use of any documents on this website.

Add your own feedback and questions here:
You are equally welcome to be positive or negative about any paper but please be polite. If you are being critical you must mention at least one specific error, otherwise your comment will be deleted as unhelpful.

comments powered by Disqus