org.aika.training

Class MetaNetwork

java.lang.Object

org.aika.training.MetaNetwork

public class MetaNetwork
extends Object

Meta-neurons and meta-synapses allow to generate new fully trained neurons based on a single training data set. The meta-network employs an inhibitory feedback loop to determine if a certain information is already known (i.e. there is already a neuron representing it) or if it is new knowledge that should be represented by a new neuron. If there is already a neuron that represents this information, then the meta-neuron will be suppressed by the feedback loop. Otherwise the meta-neuron will be activated which means a copy of the meta-neuron will be generated using only the meta information of this neuron. The meta-synapses are only added to this new neuron if the input neuron has beed active in the training data set as well. If the input neuron of the meta-synapse is another inhibitory neuron, then the resulting synapse of the new neuron is going to be connected to the activated input neuron of this inhibitory neuron.

Author:

Lukas Molzberger

Constructor Summary

Constructors

Constructor and Description
MetaNetwork()

Method Summary

Modifier and Type	Method and Description
static Neuron	initMetaNeuron(Model m, Neuron n, double bias, double metaBias, Synapse.Builder... inputs)
static void	train(Document doc)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MetaNetwork

public MetaNetwork()

Method Detail

train

public static void train(Document doc)

initMetaNeuron

public static Neuron initMetaNeuron(Model m,
                                    Neuron n,
                                    double bias,
                                    double metaBias,
                                    Synapse.Builder... inputs)