Personal tools

Network layer

From Spiking Neural Mark-up Language (SpineML)

Jump to: navigation, search

SpineML

The SpineML Network layer describes the connectivity of components using the high level concepts of populations and projections. A population contains a set of neurons which are instances of a single component model. The top level Network layer element is SpineML. It can contain only Population elements.

SpineML
Contains Description
Population [1..*] Set of one or more Population elements

Population

A Population describes a collection of neurons and a set of projections to other populations. At least one projection is required.

Population
Contains Description
Neuron [1] Neuron body definition
Projection [1..*] Set of one or more Projection elements

Neuron

The Neuron element describes a set of instances of a neuron body component model (referenced by the @url attribute). The @size and @name attributes indicate the size and name of the population respectively. Zero or more property attributes can be specified to set the initial values of StateVaraibles and Parameters of the component model. If a component model contains Parameters or StateVariables which do not have valid properties within the Neuron element then the values are assumed to be zero.

Neuron
Contains Description
Property [0..*] Set or zero or more Property Elements
@name::String The population name
@url::String URL to the neuron body component model file
@size::int The population size

Properties

Properties are used to set the value of Parameters or StateVariables for instances of a component model (this could be within a Neuron, WeightUpdate or PostSynapse). The @name attribute must reference a Parameter of StateVariable from the parents named component model. Dimensionality is inherited from the component.

A Property element references a single abstract Value element. A number of implementations of value types can be substituted wherever value is referenced.

Property
Contains Description
Delay [1] A delay element
@name::string The name of a Parameter of StateVariable from the parents named component model

Fixed Value

A fixed value indicates that the named StateVariable or Parameter for all instances of a component (within a Population, WeightUpdate or PostSynapse) have a fixed value.

FixedValue
Contains Description
@value::double Fixed value

Value List

A value list type indicates that the named StateVariable or Parameter values are set explicitly using a set of Value elements. There should be a value for each component instance (Neuron, WeightUpdate or PostSynapse). Note that value lists cannot be used where the number of weight update component instances is not fixed (i.e. for fixed probability connections).

ValueList
Contains Description
Value [1..*] Set of one or more Value elements

A Value element defines a property value for a single instance of a component. The @index attribute indicates the index position of the component instance within the set. WeightUpdate indices are interpreted depth first. i.e. For an all to all connectivity of two populations of 10 neurons, index 0 implies src neuron 0 to dst neuron 0 and index 1 implies src neuron 0 to dst neuron 1.

Value
Contains Description
@index::int Index of the component within a population, weight update or post-synapse set
@index::value Value of a property for a single instance of a component

Distribution Value

Distribution values indicates that the named StateVariable or Parameter values are set using a distribution of values. Value distributions may be either Uniform, Normal or Poisson. Each distribution may have an optional @seed attribute used to initialise the random number generation.

UniformDistribution
Contains Description
@seed::integer Optional seed value to be used for random number generation
@minimum::double Minimum value within the range of the distribution
@maximum::double Maximum value within the range of the distribution
NormalDistribution
Contains Description
@seed::integer Optional seed value to be used for random number generation
@mean::double Mean value of the distribution
@variance::double Variance of the distribution
PoissonDistribution
Contains Description
@seed::integer Optional seed value to be used for random number generation
@mean::double Mean value of the distribution

Projection

A projection links the neuron models of two populations. Connectivity is described within a projection by a synapses (each projection may have multiple synapses). The @dst_population defines the destination population name (specified within a Neuron @name element).

Projection
Contains Description
Synapse [1..*] Optional seed value to be used for random number generation
@dst_population::double Population name of the projection synapse

Synapse

A Synapse describes a specific connectivity pattern between two populations in a projection. Each synapse also contains a single weight update and post-synapse model instance which describe how neuron, synapse weight update and post-synapse model components are linked by ports.

Synapse
Contains Description
Connectivity [1] A connectivity type
WeightUpdate [1] WeightUpdate model
PostSynapse [1] PostSynapse model

Connectivity

Connectivity is used within a Synapse to describes how two populations of neurons are connected. The connectivity element is abstract, a number of implementations of connectivity types can be substituted wherever connectivity is referenced. Each connectivity type has an optional delay element. A Delay element is a special form of property which can contain only a fixed or distribution value. If no delay is specified then all delays are assumed to be 0.

OneToOneConnection

A one to one connection implies that two populations (of equal size) have a synaptic connection between source a destination neuron with the same index value. Where the projection source and destination of the projection are the same self connections are created.

OneToOneConnection
Contains Description
Delay [1] An optional synaptic delay

AllToAllConnection

An all to all connection implies that two populations (of equal size) have a synaptic connection between every source a destination neuron. Where the projection source and destination of the projection are the same self connections are created.

AllToAllConnection
Contains Description
Delay [1] An optional synaptic delay

FixedProbabilityConnection

A fixed probability connection implies that two populations (of equal size) have a synaptic connection between every source a destination neuron with a probability specified by the @probability attribute. Where the projection source and destination of the projection are the same self connections are created.

FixedProbabilityConnection
Contains Description
Delay [1] An optional synaptic delay
@probability::double [1] Probability of connectivity
@seed::integer [1] Optional seed value to use for random number generator

ConnectionList

A connection list explicitly lists synaptic connections between source a destination neurons.

ConnectionList
Contains Description
Delay [1] An optional synaptic delay
Connection [1..*] Set of one or more single connection instances

A connection element describes a single connection between a source neuron (@src_neuron) and a destination neuron (@dst_neuron). An optional delay value can be used to set a unique delay value for each synaptic connection. Where a delay value is set it will overwrite any delay value specified in the parent element.

Connection
Contains Description
@src_neuron::integer Index of the source neuron
@dst_neuron::integer Index of the destination neuron
@delay::double An optional single synaptic delay value

WeightUpdate

The WeightUpdate element describes a set of instances of a weight update component model (referenced by the @url attribute). A WeightUpdate is connect to a set of pre synaptic source neurons (from the containing projections parent population) by specifying a neuron component port (@input_src_port attribute) and weight update component port (@input_dst_port attribute) to link the components. Feedback from a post synaptic neuron to the weight update (for implmenting learning mechanisms) is possible by using the @feedback_src_port attribute to specify the post synaptic neuron component port and @feedback_dst_port attribute to specify the synaptic component port. Zero or more property attributes can be specified to set the initial values of StateVaraibles and Parameters of the weight update component model. If a component model contains Parameters or StateVariables which do not have valid properties within the Neuron element then the values are assumed to be zero. No size is specified for a weight update component. Size is determined by the source and destination populations sizes and the connectivity pattern used to connect them.

WeightUpdate
Contains Description
Property [0..*] Set or zero or more Property elements
@name::String The population name
@url::String URL to the neuron body component model file
@input_src_port::string Source port of pre synaptic neuron component
@input_dst_port::string Destination port of weight update component for pre-synaptic input
@feedback_src_port::string Optional destination port of post-synapse component to communicate weight update data for learning
@feedback_dst_port::string Optional source port of weight update component to communicate feedback to post-synaptic neuron for learning

PostSynapse

The PostSynapse element describes a set of instances of a post-synapse component model (referenced by the @url attribute). A PostSynapse is connect to a set of synapes by specifying a weight update component port (@input_src_port attribute) and the post-synapse component port (@input_dst_port attribute) to link the components. The post-synapse connects to a set of post synaptic neurons by using the @output_src_port attribute to specify the post-synapse component port and @output_dst_port attribute to specify the post synaptic neuron component port. Zero or more property attributes can be specified to set the initial values of StateVaraibles and Parameters of the post-synapse component model. If a component model contains Parameters or StateVariables which do not have valid properties within the Neuron element then the values are assumed to be zero. No size is specified for a post-synapse component. Size is determined by the destination populations size.

PostSynapse
Contains Description
Property [0..*] Set or zero or more Property elements
@name::String The population name
@url::String URL to the neuron body component model file
@input_src_port::string Source port of weight update component
@input_dst_port::string Destination port of post-synapse component for synaptic input
@output_src_port::string Source port of post-synpase component
@output_dst_port::string Destination port of post-synaptic neuron for post synaptic output