Human TSP Solver

Well, it has been approximately three weeks since the inception of the (crazy) idea for a human-driven TSP solving web application. I've you've haven't been tracking the progress see the inception, the consideration of user experience, getting up and running, and the discovery of a related project.

I am happy to reveal that I have released a version of the web application to private alpha testing! Given that the audience of this blog is about 50 people, (private by anyones standards) type in the title of this post .com to have a look, but only if you promise to give me some feedback!

The current state of the application is stable and offers the core functionality outlined in the projects inception, specifically: the capturing of user contributions toward a crowdsourced solution to the travelling salesman problem. I used Ruby on Rails as the web framework and Java Applets to capture and display user contributions.

After having the site up for a day, it is clear to me that I am primarily interested in the aggregate data, what it means, and the ways in which it may be harnessed or applied. Naturally, there is plenty of room for additional features, specifically focused around how contributions are provided by users (various games), although additional features concerning the aggregate contributions are harder to brainstorm. I think this is simply because I don't know what to expect. I don't know what it means to have tens or hundreds of thousands of partial solutions to TSP's. What I am sure of is that such information is far more useful than simply solving small instances of the general problem.

While pondering this question and preparing for a presentation of the project to my research group last week, I came across a series of papers from the Journal of Problem Solving (vol 1, iss 1) dedicated to the investigation of human-based TSP solving (credit to the Wikipedia TSP article). I plan on reading these in depth, although a first pass highlights some interesting and relevant findings regarding the low cognitive load and high performance of humans on small instances.

Anyway, it feels good to have a usable alpha version out the door, and exciting to consider the array of potential features that may be integrated into the next iteration.

Thanks to my my early alpha testers, your feedback has been invaluable!