Eating an octopus tentacle
in Hakodate, Japan, May 2006

Mark J. Nelson

Georgia Tech College of Computing
mnelson@cc (.gatech.edu)

I'm a PhD student in computer science working in artificial intelligence.

My dissertation work-in-progress is on automated support for game design. This has a number of targets, from enabling game-design experts to prototype and refine design ideas more quickly; to enabling novices to build interesting small games; to even allowing games or parts of games to be automatically generated.

I've done some work on drama management, and longer ago on the periphery of computer music. Lurking in the todo pile is some as-yet-unpublished stuff on machine learning, mainly local regression methods and regression forests.

I'm working in Michael Mateas's Expressive Intelligence Studio (EIS), now located at the University of California, Santa Cruz. I've also collaborated with Charles Isbell's pfunk research group, and Alex Gray's FASTlab.

You might also be interested in an index of newsgames I maintain.


Publications

Automated Support for Game Design

We feel that there's huge potential for automated methods to assist in game design, from helping experts prototype and design games faster; to helping novices design games at all; to allowing games to dynamically customize themselves. Some previous research has looked at automatically designing variations on certain classes of games (mainly chess variants), but we'd like to look at the whole problem: game mechanics, concrete audiovisual representation, real-world references, and player input. In addition, we'd like to stress the interactive support aspect: spitting out a variant on a game is not the only (or even, necessarily, the most useful) application of automated game design. The papers below are preliminary work outlining some of the ideas in the context of reasoning about a game's real-world references. A paper on reasoning about a game's mechanics should be forthcoming shortly.

Declarative Optimization-Based Drama Management (DODM)

Drama managers watch a game (or other interactive experience) as it progresses, and intervene when necessary to keep the experience interesting and in line with an author's goals. DODM is a particular approach in which an author declaratively specifies what it would mean for the experience to go well, along with an abstraction of the story and a set of interventions the system can make. The system then optimizes its interventions according to the given criteria.

The basic approach was proposed by Joe Bates in 1992 and developed by Peter Weyhrauch in 1997. We've since made a number of modifications, both to the conceptual formulation and to the technical implementation.

MIDI Performance

I evaluated MIDI performance as part of previous work (2003–2004) with Belinda Thom at Harvey Mudd College. We were trying to discover interesting things about jazz improvisation, but got sidetracked into measuring the performance of our MIDI equipment, since we needed to be able to put error bars on the data. You may also want to take a look at the project's site, which has code, circuit diagrams, and everything else you need to run your own tests, as well as the detailed results of our tests.


This information was current as of: May 2008
John Balance is dead. | Personal finance stuff.