Home * People * Daniel Walker

Daniel Walker,
an American computer scientist and Linux kernel engineer. He holds a B.Sc. from the Jack Baskin School of Engineering at University of California, Santa Cruz (UCSC) in 2004, where he worked with Robert Levinson on the learning chess program Morph. He implemented a system to save and restore the learned weights in the neural network (NN) - so if the performance level of the NN drops, a known good copy of the NN is restored to continue playing. Morph's learning rate was further adjusted to be roughly inversely proportional to its chess rating. Similar to adjustment simulated annealing, Morph's weights change less the better Morph plays. Daniel Walker also worked on taking better advantage of the neighborhood representation, expanding the specifity of weights used by the NN without increasings its size, i.e. rather than exploring the same neighborhood combining rule for all squares, considering piece color, and type and/or location [1].

Selected Publications

[2]

External Links


References

  1. ^ Daniel Walker, Robert Levinson (2004). The MORPH Project in 2004. ICGA Journal, Vol. 27, No. 4
  2. ^ dblp: D. Walker

What links here?


Up one level