Date: Mon, 01 Mar 1999 15:53:09 +1030 From: Luke Pellen <luke-AT-seol.net.au> Subject: Just so you know what I'm up to... How does an ANN [artificial neural network] work? ------------------------------------------------- An ANN is an implementation of "connectionist" architecture. A connectionist architecture seeks to loosely emulate the workings of a biological brain and is characterized by having a large number of very simple neuron-like processing elements, a large number of weighted connections between these elements, parallel distributed control, and an emphasis on learning internal representations automatically. The weights connecting each neuron encode the knowledge of the network. A simulated neuron is a node connected to other nodes via links that approximate to axon-synapse-dendrite connections. Each link is associated with a weight. The connecting weight multiplied by the neuron's output determines the nature and strength of one node's influence on another: a large positive weight corresponds to strong excitation, and a small negative weight corresponds to weak inhibition. How does Octavius work? ----------------------- Octavius is classified as a feed-forward network: he has 768 input nodes, 20 hidden nodes, and 16 output nodes. Each input node is linked to each hidden node, and each hidden node is linked to each output node. This gives Octavius a total of 804 nodes and 15680 individually weighted connections. A chess position is encoded as binary and fed into the input neurons. The reason for 768 input nodes is that there are 64 squares on a chess board, and each square has 12 possible states [my pawn, enemy pawn, my bishop, enemy bishop etc.]. An empty square is simply represented as 12 zeros. The hidden layer consists of 20 nodes; this number is purely arbitrary and subject to change pending further experimentation. The hidden layer is where the bulk of the processing is performed. The output layer consists of 16 nodes which, again, is an arbitrary figure essentially allowing Octavius 16 levels of judgment. The raw values of these output nodes are totaled to give a numerical positional evaluation. How does Octavius learn? ------------------------ A human who learns how to play chess begins with the basics: the movements of the pieces, the relative values of the pieces, threats, traps and discovered checks etc. Octavius attempts to gain an understanding of chess through a positional analysis of master and grand master games. His training is based on the assumption that any position reached during such games must be positionally superior to any alternatively available position during that game. It must be stressed that Octavius has absolutely no hard coded knowledge of chess. My theory is that through exposure to master and grand master games Octavius will be able to deduce the rules and tactics of chess heuristically via positional interpolation. This top-down method is precisely the reverse of the human bottom-up approach. His performance to date has been close to that of a very bad human chess player. Here is one of his best games against me (he starts quite well, but deteriorates as the game progresses): White: Luke Pellen Black: Octavius 1. e4 d5 2. exd5 e5 3. Nc3 Nf6 4. d3 e4 5. dxe4 Bb4 6. Bd2 Bxc3 7. Bxc3 Nxe4 8. Nf3 Nf6 9. Be2 Nxd5 10. O-O Ne3 (nice try Octavius... ) 11. fxe3 Bf5 12. Nd4 c5 13. Nxf5 c4 14. Nxg7+ Ke7 15. Bh5 f5 16. Nxf5+ Ke6 17. Qg4 b5 18. Rad1 b4 19. Ng7+ Ke7 20. Rf7# {White mates} 1-0 Luke Pellen e-mail: luke-AT-seol.net.au ICQ#: 25510475 -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- For in and out, above, about, below, 'Tis nothing but a Magic Shadow-show Play'd in a Box whose Candle is the Sun, Round which we Phantom Figures come and go. - The Rubaiyat Of Omar Khayyam -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Chaotic Pearl: http://members.tripod.com/~vidagnosis/journal.html This random quotation was generated by SIGGEN... SIGGEN is an e-mail signature generator programmed by Luke Pellen
Display software: ArchTracker © Malgosia Askanas, 2000-2005