Are Self-Organizing Maps Just an Exercise in Relativism?

The key to unlocking the power of a self-organizing map seems to be in this phrase by Diego Vicente:

…instead of a grid we declare a circular array of neurons, each node will only be conscious of the neurons in front of and behind it…

He offers the example of Uruguay

traversing 734 Uruguay cities only 7.5% longer than the optimal in less than 25 seconds

In other words, each node should dispense with attempts to measure on an absolutist grid and instead calculate its own position relative to other nodes in the immediate vicinity. Like modadism, but nodadism. Also like the difference between racing single-track on a mountain bike (stay ahead of the person behind, get in front of person ahead) and racing road bikes on a highway (pre-calculate best times of pursuit, rest and attack).

Diego refers to a node’s immediate vicinity as “moderate exploitation of the local minima of each part” of a larger grid. That makes perfect sense for anyone familiar with navigating by asking around. Ask a local which way to the closest next town, if you can find a trusted local. Don’t bother asking them for a way to towns they never see, and be able to recognize the difference.

The more I research flaws in AI security the more the world bifurcates into the grey and ill-defined transition from relative to absolute models of authentication and authorization. In between there are many exploits to be found.

The problem set here is called the National Travelling Salesman by mathematicians. Of course in security terms we should think of this as drone routes to destroy privacy (gather knowledge, if you prefer that angle) or an estimation of resources for a comprehensive integrity attack plan (defense, if you prefer that angle).

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.