Continuous attractor neural field — the core DFT building block. More...

#include <neural_field.h>

Inheritance diagram for dnf_composer::element::NeuralField:

Collaboration diagram for dnf_composer::element::NeuralField:

Public Member Functions
std::shared_ptr< Element >	clone () const override

std::vector< NeuralFieldBump >	getBumps () const
	Return all currently detected above-threshold bumps.

bool	getComputeStateMetrics () const

double	getHighestActivation () const

double	getLowestActivation () const

NeuralFieldParameters	getParameters () const

std::shared_ptr< Kernel >	getSelfExcitationKernel () const
	Return the registered self-excitation kernel, if any.

double	getStabilityThreshold () const

void	init () override
	Initialize the element (called once before the simulation loop).

bool	isStable () const

	NeuralField (const ElementCommonParameters &elementCommonParameters, const NeuralFieldParameters &parameters)
	Construct a neural field.

void	setComputeStateMetrics (bool enable)
	Enable or disable per-step state-metric computation (stability, bumps, min/max). The default (enabled) path uses a single fused O(N) pass and is fast enough for normal use. Disable only as an advanced micro-optimisation for headless batch runs where bump data and stability checks are never needed. Default: true (full state tracking enabled).

void	setParameters (const NeuralFieldParameters &parameters)

void	setThresholdForStability (const double threshold)
	Set the stability convergence threshold.

void	step (double t, double deltaT) override
	Advance the element by one time step.

std::string	toString () const override

Public Member Functions inherited from dnf_composer::element::Element
virtual void	addInput (const std::shared_ptr< Element > &inputElement, const std::string &inputComponent="output")
	Register `inputElement` as an upstream source for this element.

void	buildInputCache ()
	Cache raw pointers to input component data. Call after all element init()s complete.

virtual void	changeDimensions (const ElementDimensions &newDimensions)
	Resize all components to `newDimensions` and re-initialize.

void	close ()

	Element (const ElementCommonParameters &parameters)
	Construct an element with the given common parameters.

std::vector< double >	getComponent (const std::string &componentName)
	Return a copy of the named component vector.

std::vector< std::string >	getComponentList () const

std::vector< double > *	getComponentPtr (const std::string &componentName)

const std::unordered_map< std::string, std::vector< double > > *	getComponents () const
	Return a read-only pointer to the full components map.

ElementCommonParameters	getElementCommonParameters () const

std::vector< std::shared_ptr< Element > >	getInputs ()

std::unordered_map< std::shared_ptr< Element >, std::string >	getInputsAndComponents ()
	Return all inputs mapped to the component name they expose.

ElementLabel	getLabel () const

int	getMaxSpatialDimension () const

std::vector< std::shared_ptr< Element > >	getOutputs ()

int	getSize () const
	Return the number of spatial samples (size = round(x_max / d_x)).

double	getStepSize () const
	Return the spatial resolution (d_x).

int	getUniqueIdentifier () const

std::string	getUniqueName () const

bool	hasInput () const

bool	hasInput (const std::string &inputElementName, const std::string &inputComponent)

bool	hasInput (int inputElementId, const std::string &inputComponent)

bool	hasOutput () const

bool	hasOutput (const std::string &outputElementName, const std::string &outputComponent)

bool	hasOutput (int outputElementId, const std::string &outputComponent)

void	print () const

void	removeInput (const std::string &inputElementId)

void	removeInput (int uniqueId)

void	removeInputs ()

void	removeOutput (const std::string &outputElementId)
	Deregister this element as an input of `outputElementId`.

void	removeOutput (int uniqueId)
	Deregister this element as an input of the element with `uniqueId`.

void	removeOutputs ()

void	setUniqueName (const std::string &name)

void	updateInput ()
	Pull data from all registered input elements into this element's components.

virtual	~Element ()=default

Protected Member Functions
void	calculateActivation (double t, double deltaT)

void	calculateOutput ()

void	updateBumps (double deltaT)

void	updateState (double deltaT)

Protected Attributes
NeuralFieldParameters	parameters
	Dynamics parameters (tau, h, activation function).

NeuralFieldState	state
	Runtime state (bumps, stability, min/max).

Protected Attributes inherited from dnf_composer::element::Element
ElementCommonParameters	commonParameters
	Name, label, and spatial dimensions.

std::unordered_map< std::string, std::vector< double > >	components
	Named data arrays (e.g. "output").

std::unordered_map< std::shared_ptr< Element >, std::string >	inputs
	Upstream elements and the component they expose.

std::unordered_map< std::shared_ptr< Element >, std::string >	outputs
	Downstream elements that read this element's output.

Detailed Description

Continuous attractor neural field — the core DFT building block.

Implements a one-dimensional neural field with Amari-type dynamics:

tau * du/dt = -u + h + s(x, t) + (w * f(u))(x)

where u is activation, h the resting level, s the summed external input, and w the lateral interaction kernel convolved with the output f(u) (determined by the activation function).

Constructor & Destructor Documentation

◆ NeuralField()

dnf_composer::element::NeuralField::NeuralField	(	const ElementCommonParameters &	elementCommonParameters,
		const NeuralFieldParameters &	parameters
	)

Construct a neural field.

Parameters

elementCommonParameters	Name, label, and spatial dimensions.
parameters	Field dynamics parameters.

Member Function Documentation

◆ calculateActivation()

void dnf_composer::element::NeuralField::calculateActivation	(	double	t,
		double	deltaT
	)

protected

Here is the caller graph for this function:

◆ calculateOutput()

void dnf_composer::element::NeuralField::calculateOutput ( )

protected

Here is the caller graph for this function:

◆ clone()

std::shared_ptr< Element > dnf_composer::element::NeuralField::clone ( ) const

overridevirtual

Implements dnf_composer::element::Element.

◆ getBumps()

std::vector< NeuralFieldBump > dnf_composer::element::NeuralField::getBumps ( ) const

inline

Return all currently detected above-threshold bumps.

◆ getComputeStateMetrics()

bool dnf_composer::element::NeuralField::getComputeStateMetrics ( ) const

inline

◆ getHighestActivation()

double dnf_composer::element::NeuralField::getHighestActivation ( ) const

inline

◆ getLowestActivation()

double dnf_composer::element::NeuralField::getLowestActivation ( ) const

inline

◆ getParameters()

NeuralFieldParameters dnf_composer::element::NeuralField::getParameters ( ) const

◆ getSelfExcitationKernel()

std::shared_ptr< Kernel > dnf_composer::element::NeuralField::getSelfExcitationKernel ( ) const

Return the registered self-excitation kernel, if any.

◆ getStabilityThreshold()

double dnf_composer::element::NeuralField::getStabilityThreshold ( ) const

inline

◆ init()

void dnf_composer::element::NeuralField::init ( )

overridevirtual

Initialize the element (called once before the simulation loop).

Implements dnf_composer::element::Element.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ isStable()

bool dnf_composer::element::NeuralField::isStable ( ) const

◆ setComputeStateMetrics()

void dnf_composer::element::NeuralField::setComputeStateMetrics ( bool enable )

inline

Enable or disable per-step state-metric computation (stability, bumps, min/max). The default (enabled) path uses a single fused O(N) pass and is fast enough for normal use. Disable only as an advanced micro-optimisation for headless batch runs where bump data and stability checks are never needed. Default: true (full state tracking enabled).

◆ setParameters()

void dnf_composer::element::NeuralField::setParameters ( const NeuralFieldParameters & parameters )

Here is the call graph for this function:

◆ setThresholdForStability()

void dnf_composer::element::NeuralField::setThresholdForStability ( const double threshold )

inline

Set the stability convergence threshold.

Parameters

threshold New threshold value (default 0.895).

◆ step()

void dnf_composer::element::NeuralField::step	(	double	t,
		double	deltaT
	)

overridevirtual

Advance the element by one time step.

Parameters

t	Current simulation time.
deltaT	Integration step size.

Implements dnf_composer::element::Element.

Here is the call graph for this function:

◆ toString()

std::string dnf_composer::element::NeuralField::toString ( ) const

overridevirtual

Implements dnf_composer::element::Element.

Here is the call graph for this function:

◆ updateBumps()

void dnf_composer::element::NeuralField::updateBumps ( double deltaT )

protected

Here is the caller graph for this function:

◆ updateState()

void dnf_composer::element::NeuralField::updateState ( double deltaT )

protected

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

◆ parameters

NeuralFieldParameters dnf_composer::element::NeuralField::parameters

protected

Dynamics parameters (tau, h, activation function).

◆ state

NeuralFieldState dnf_composer::element::NeuralField::state

protected

Runtime state (bumps, stability, min/max).

The documentation for this class was generated from the following files:

include/elements/neural_field.h
src/elements/neural_field.cpp

Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ NeuralField()

Member Function Documentation

◆ calculateActivation()

◆ calculateOutput()

◆ clone()

◆ getBumps()

◆ getComputeStateMetrics()

◆ getHighestActivation()

◆ getLowestActivation()

◆ getParameters()

◆ getSelfExcitationKernel()

◆ getStabilityThreshold()

◆ init()

◆ isStable()

◆ setComputeStateMetrics()

◆ setParameters()

◆ setThresholdForStability()

◆ step()

◆ toString()

◆ updateBumps()

◆ updateState()

Member Data Documentation

◆ parameters

◆ state