Thursday, June 03, 2004

Greedy algorithm could help ranking physical phenomena.

No I do not mean this type of greedy algorithm, where you shut down plants to hike up the prices. Rather, it is a type of algorithm used in networks that is pretty fast and is based on initial sensible solutions. As I was reading the very interesting list of publications by Mark Newmann, I realized I was looking at the other part of the sorting/ranking algorithm that I needed. When one has performed a pairwise estimation of the distance between different objects of a set, what is the best way to plot all these relationships. Sure you could try to plot the correspondance matrix but it would likely confuse everybody but yourself. The article on this fast algorithm devised a way to build a fast ranking of different elements with each other. Let me take an example: let us imagine that you have different photos of a two-phase flow experiment and you want to categorize them in different sets called "bubbly flow", "annular flow". You first compute the "distance" between each pictures. If you have n photos, it means you now have n^2 distances. You then use the algorithm of Newmann highlighted above on the set of distances and you should be able to have different categories of photos representing different catagories of flows. If it works, the interesting thing is that it would be an automated process as opposed to having an human operator qualifying the flow regime.

No comments:

Printfriendly