David E. Moriarty

From Chessprogramming wiki
Jump to: navigation, search

Home * People * David E. Moriarty

David Eric Moriarty,
an American computer scientist an Ph.D. alumni from University of Texas at Austin [1]. During the 90s, along with his advisor, Risto Miikkulainen, David E. Moriarty worked on reinforcement learning by Symbiotic, Adaptive NeuroEvolution dubbed SANE, also topic of his Ph.D. thesis [2].

SANE

SANE (Symbiotic, Adaptive NeuroEvolution) evolves neural networks with genetic algorithms for sequential decision tasks, also applied to the games of Othello and Go. SANE selects a population of hidden neurons of a "vanilla" three-layer feed-forward neural network with the connections and weights in both directions, performing following basic steps in one generation [3]:

1. Clear all fitness values from each neuron
2.   Select neurons randomly from the population
3.   Create a neural network from the selected neurons
4.   Evaluate the network in the given task
5.   Add the network's score to each selected neuron's fitness variable
6. Repeat steps 2-5 a sufficient number of times
7. Get each neuron's average fitness score by dividing its total fitness values by the number of networks in which it was implemented
8. Perform crossover operations on the population based on the average fitness value of each neuron

For Go and Othello, the input layer sees the board configuration, while the output layer indicates the goodness of each possible move by an output neuron associated with each space or point of the board. However the research was conducted a few years before the MCTS revolution appeared in computer Go, not to mention the deep learning breakthrough. In the alpha-beta search of the Othello experiment, the neural network orders the moves, and further controls whether moves are pruned forward [4].

Selected Publications

[5]

External Links

References

Up one Level