org.aika.neuron

Class Synapse

java.lang.Object

org.aika.neuron.Synapse

All Implemented Interfaces:

Writable

public class Synapse
extends Object
implements Writable

The Synapse class connects two neurons with each other. When propagating an activation signal, the weight of the synapse is multiplied with the activation value of the input neurons activation. The result is then added to the output neurons weighted sum to compute the output activation value. In contrast to a conventional neural network synapses in Aika do not just propagate the activation value from one neuron to the next, but also structural information like the text range and the relational id (e.g. word position) and also the interpretation to which the input activation belongs. To determine in which way the structural information should be propagated synapses in Aika possess a few more properties.

The properties relativeRid and absoluteRid determine either the relative difference between the rid of the input activation and the rid of the output activation or require a fixed rid as input.

The properties range match, range mapping and range output manipulate the ranges. The range match determines whether the input range begin or end is required to be equal, greater than or less than the range of the output activation. The range mapping can be used to map for example an input range end to an output range begin. Usually this simply maps begin to begin and end to end. The range output property is a boolean flag which determines whether the input range should be propagated to the output activation.

Furthermore, the property isRecurrent specifies if this synapse is a recurrent feedback link. Recurrent feedback links can be either negative or positive depending on the weight of the synapse. Recurrent feedback links kind of allow to use future information as inputs of a current neuron. Aika allows this by making assumptions about the recurrent input neuron. The class SearchNode modifies these assumptions until the best interpretation for this document is found.

Author:

Lukas Molzberger

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	Synapse.Builder The Builder class is just a helper class which is used to initialize a neuron.
static class	Synapse.Key

Field Summary

Fields

Modifier and Type	Field and Description
double	bias
double	biasDelta
Neuron	input
static Comparator<Synapse>	INPUT_SYNAPSE_COMP
Provider<InputNode>	inputNode
boolean	isConjunction The synapse is stored either in the input neuron or the output neuron depending on whether it is a conjunctive or disjunctive synapse.
Synapse.Key	key
MetaSynapse	meta
Neuron	output
static Comparator<Synapse>	OUTPUT_SYNAPSE_COMP
boolean	toBeDeleted
double	weight The weight of this synapse.
double	weightDelta The weight delta of this synapse.

Constructor Summary

Constructors

Constructor and Description
Synapse()
Synapse(Neuron input, Neuron output)
Synapse(Neuron input, Neuron output, Synapse.Key key)

Method Summary

Modifier and Type	Method and Description
static Synapse	createOrLookup(Synapse.Key synapseKey, Neuron inputNeuron, Neuron outputNeuron)
boolean	exists()
boolean	isConjunction(boolean v)
boolean	isNegative()
void	link()
static Synapse.Key	lookupKey(Synapse.Key k)
static Synapse	read(DataInput in, Model m)
void	readFields(DataInput in, Model m) Deserialize the fields of this object from in.
String	toString()
void	unlink()
void	write(DataOutput out) Serialize the fields of this object to out.

Methods inherited from class java.lang.Object

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

Field Detail

INPUT_SYNAPSE_COMP

public static final Comparator<Synapse> INPUT_SYNAPSE_COMP

OUTPUT_SYNAPSE_COMP

public static final Comparator<Synapse> OUTPUT_SYNAPSE_COMP

input

public Neuron input

output

public Neuron output

inputNode

public Provider<InputNode> inputNode

key

public Synapse.Key key

weight

public double weight

The weight of this synapse.

weightDelta

public double weightDelta

The weight delta of this synapse. The converter will use it to compute few internal parameters and then update the weight variable.

bias

public double bias

biasDelta

public double biasDelta

toBeDeleted

public boolean toBeDeleted

isConjunction

public boolean isConjunction

The synapse is stored either in the input neuron or the output neuron depending on whether it is a conjunctive or disjunctive synapse.

meta

public MetaSynapse meta

Constructor Detail

Synapse

public Synapse()

Synapse

public Synapse(Neuron input,
               Neuron output)

Synapse

public Synapse(Neuron input,
               Neuron output,
               Synapse.Key key)

Method Detail

link

public void link()

unlink

public void unlink()

exists

public boolean exists()

isConjunction

public boolean isConjunction(boolean v)

isNegative

public boolean isNegative()

toString

public String toString()

Overrides:

toString in class Object

write

public void write(DataOutput out)
           throws IOException

Description copied from interface: Writable

Serialize the fields of this object to out.

Specified by:

write in interface Writable

Parameters:

out - DataOuput to serialize this object into.

Throws:

IOException

readFields

public void readFields(DataInput in,
                       Model m)
                throws IOException

Description copied from interface: Writable

Deserialize the fields of this object from in.

For efficiency, implementations should attempt to re-use storage in the existing object where possible.

Specified by:

readFields in interface Writable

Parameters:

in - DataInput to deseriablize this object from.

Throws:

IOException

read

public static Synapse read(DataInput in,
                           Model m)
                    throws IOException

Throws:

IOException

lookupKey

public static Synapse.Key lookupKey(Synapse.Key k)

createOrLookup

public static Synapse createOrLookup(Synapse.Key synapseKey,
                                     Neuron inputNeuron,
                                     Neuron outputNeuron)