Gone To Santa Fe

The Move

So we fairly quickly changed gears, worked like crazy getting the house ready for closing on November 14 and for moving on the 20th.  Whew!  Exhausting!   But we were soon on the road to Santa Fe, picked up Buster the Cat who flew into Albuquerque rather than drive with us, and were residents here by the end of the month.  With our name freshly applied to our mail box, we knew we had arrived!

What Next?

So, once landed, what to do? We decided to try "early retirement".  Indeed, we had previously planned to retire at exactly the same month as the layoff, but had stayed on due to the down market.  But we also wanted to stay  involved with the various technical pursuits I had become involved with at the Santa Fe Institute. To start with (after getting the home computer nook on the air), I got quite involved with FRIAM, the Friday AM complexity gang, looking into interesting pursuits.  This included getting a HostGo web hosting service and migrating my backspaces.net environment there.  I've also gotten a new domain for a potential consultancy: ComplexityWorkshop.com.

Office

While getting settled into the local complexity community, I had a chance to acquire an office with RedFish, a successful consultancy run by my friend Stephen Guerin.  It's in a lovely old adobe in down town Santa Fe.

O'Reilly

My first plunge was to write a few new blogs for O'Reilly, including Domain Maddness and Beyond Crypto.  Domain Maddness chronicals my plunge into choosing a domain name, including pushing all of the 234,937 words in /usr/share/dict/words through Whois. Beyond Crypto presents a brief discussion of "adaptive security" methods presented at the Santa Fe Institute's Adaptive Defense of Computing Networks workshop.  Blogging turns out to be a great way to meet a wide set of computer folks.

Value Network Project

The ValueNet project (SFI Business Network) decided on its second "extreme programming" session right about then.  This is a supply chain simulation and exploration project including Argonne, Los Alamos and Sandia national laboratories, along with myself and other independent researchers. This session took as its focus two areas: visualization and mesh supply networks.  Due to my being in Santa Fe, while the session took place in Chicago, it was decided I should modify an earlier version of the RePast simulation to allow for more complex networks of suppliers.  We would then have two independent implementations to validate our results.

The Beer Game simulation we used typically showes huge volatility in inventory, order rate, and costs.  The order rate graph, below left, shows continuous extreme volatility. Our earlier work had shown promising results: that increased visibility within the supply chain can reduce this volatility somewhat. Our second session focused on migrating from a single chain of Factory - Distributer - Wholesaler - Retailer (left) to a more complicated mesh or network graph (right). The result was quite dramatic.  The second graph below right, shows that the volatility is initially large but dampens to steady state over time.

Our group's next step is to present at this year's Lake Arrowhead Conference on Human Complex Systems 2003 (March 19-23).

Peer Net

After several conversations with local consultants and contractors in the scientific area, I decided we badly needed a small set of tutorials giving simplified but detailed examples of 4-6 software techniques used in complex systems.  These could be used in conference tutorial tracks, as material for a book, or introductory material for consulting jobs. The first two I decided to do are simulations, one using NetLogo, the next using RePast, both popular, publicly available, Java-based simulation packages.

The NetLogo program is a very simple peer network simulation.  It provides "knobs" to select how often peers spring to life, how long they live, how far they "reach" for other peers, and how many  links peers have to other peers.  The simulation shows how connected the peer network is, showing the disconnected sub-networks in different colors.

One of the great advantages of NetLogo is that the simulations can be packaged into an applet for viewing and running online!  I've put the peer network simulation online for you to play with.  Interestingly enough, even though limited in radius and lifetime, the peer network exhibits the classic closure parameters found by Erdos and Renyi, where the number of edges is roughtly one per node when the closed network emerges.  This is a bit surprising considering the limits imposed by radius-limited edge lengths.

Ant Colony Optimization

My most recent exploration is a RePast simulation of an Ant Colony Optimization algorithm for the Travelling Salesman problem.  The ACO technique is a modern heuristic methodology for finding "good enough" solutions to intractable problems in reasonable time. The RePast simulation is meant to be simple enough for teaching purposes, yet effective enough to be useful in field situations like Supply Chain Logistics and similar areas.

Here is the simulation running under RePast after 36 steps.  The control strip shows the tick count, along with go, step, repeat, stop, pause, reset, file, help, and quit buttons.  The parameter settings on the right panel can be used to change the number of nodes and other run time parameters of the simulations.

Note that the current tour is not optimal, the topmost red node (denoting a node at which ants start) shows two ineffeciencies in its vicinity.  In order to understand the routes taken by the ants, you can click on the "PaintTraversals" button which has all the ants paint the paths they took.  Using this produces this 37th step:

We can "debug" this way, noticing that the optimum paths were not even tried.  The solution to this problem is the additional "Candidate List" optimization discussed in the literatuture.

Although RePast does not have an Applet feature like NetLogo, it is fairly easy to run locally as a Java application within the RePast framework.  Here is an ACO.tar.gz file containing the ACO source code.