Home * People * Jens Christensen
pic_jens.jpg

Jens Christensen,
an American computer scientist with a Ph.D. in 1991 from Stanford University on automatic abstraction in planning [1] . Along with Richard Korf, he elaborated on a Unified Theory of Heuristic Evaluation functions and Its Applications to Learning on optimization of feature weights by Hill climbing methods and linear regression. In one experiment, a set of relative weights for different chess pieces was automatically learned.
Jens Christensen [2]

Experiments with Chess

Excerpt from A Unified Theory of Heuristic Evaluation functions and Its Applications to Learning. [3] :
As a serious test, we chose the game of chess and a simple evaluation function consisting only of material advantage. The experiment was to see if the learning program would approximate the classically accepted weights for the pieces: 9 for the queen, 5 for the rook, 3 for the bishop, 3 for the knight, and 1 for the pawn.

The chess program was implemented using a two-ply mini-max search with alpha-beta pruning and quiescence. 1400 half moves were made between each regression. If neither side won during a game it was stopped after 100 half-moves and a new game was started. For purposes of the experiment, a win was assigned one more than the total initial material value, and the individual piece values were rounded off to the nearest 0.5. The pieces stabilized at: Queen, 8.0; rook, 4.0; bishop, 4.0; knight, 3.0; pawn 2.0. ...

Selected Publications

[4]

External Links


References

  1. ^ Jens Christensen (1991). Automatic Abstraction in Planning. Ph.D. thesis, Stanford University
  2. ^ Thane E. Plambeck bio - PostModern Computing Technologies, Inc. (1991-1995) - Jens Christensen bio, photo
  3. ^ Jens Christensen, Richard Korf (1986). A Unified Theory of Heuristic Evaluation functions and Its Applications to Learning. Proceedings of the AAAI-86, pp. 148-152, 3.3. Experiments with Chess, pp. 151, pdf
  4. ^ DBLP: Jens Christensen

What links here?


Up one level