Class implementing a synapse model for e-prop plasticity with additional biological features. More...

#include <eprop_synapse.h>

Inheritance diagram for nest::eprop_synapse< targetidentifierT >:

Collaboration diagram for nest::eprop_synapse< targetidentifierT >:

Classes
class	ConnTestDummyNode
	Dummy node for testing the connection. More...

Public Types
typedef EpropSynapseCommonProperties	CommonPropertiesType
	Type of the common synapse properties.

typedef Connection< targetidentifierT >	ConnectionBase
	Type of the connection base.

Public Member Functions
	eprop_synapse ()
	Default constructor.

	~eprop_synapse ()
	Destructor.

	eprop_synapse (const eprop_synapse &)
	Parameterized copy constructor.

eprop_synapse &	operator= (const eprop_synapse &)
	Assignment operator.

	eprop_synapse (eprop_synapse &&)
	Move constructor.

eprop_synapse &	operator= (eprop_synapse &&)
	Move assignment operator.

void	get_status (Dictionary &d) const
	Get parameter dictionary.

void	set_status (const Dictionary &d, ConnectorModel &cm)
	Update values in parameter dictionary.

bool	send (Event &e, size_t thread, const EpropSynapseCommonProperties &cp)
	Send the spike event.

void	check_connection (Node &s, Node &t, size_t receptor_type, const CommonPropertiesType &cp)
	Check if the target accepts the event and receptor type requested by the sender.

void	set_weight (const double w)
	Set the synaptic weight to the provided value.

void	delete_optimizer ()
	Delete optimizer.

double	get_delay () const
	Return the delay of the connection in ms.

long	get_delay_steps () const
	Return the delay of the connection in steps.

size_t	get_rport () const

Node *	get_target (const size_t tid) const

Public Member Functions inherited from nest::Connection< targetidentifierT >
	Connection ()

	Connection (const Connection< targetidentifierT > &rhs)=default

Connection &	operator= (const Connection< targetidentifierT > &rhs)=default

std::unique_ptr< SecondaryEvent >	get_secondary_event ()
	Get a pointer to an instance of a SecondaryEvent if this connection supports secondary events.

void	get_status (Dictionary &d) const
	Get all properties of this connection and put them into a dictionary.

void	set_status (const Dictionary &d, ConnectorModel &cm)
	Set properties of this connection from the values given in dictionary.

void	check_synapse_params (const Dictionary &d) const
	Check syn_spec dictionary for parameters that are not allowed with the given connection.

void	calibrate (const TimeConverter &)
	Calibrate the delay of this connection to the desired resolution.

double	get_delay () const
	Return the delay of the connection in ms.

long	get_delay_steps () const
	Return the delay of the connection in steps.

void	set_delay (const double delay)
	Set the delay of the connection.

void	set_delay_steps (const long delay)
	Set the delay of the connection in steps.

void	set_syn_id (synindex syn_id)
	Set the synapse id of the connection.

synindex	get_syn_id () const
	Get the synapse id of the connection.

long	get_label () const

void	trigger_update_weight (const size_t, const std::vector< spikecounter > &, const double, const CommonSynapseProperties &)
	Triggers an update of a synaptic weight.

Node *	get_target (const size_t tid) const

size_t	get_rport () const

void	set_source_has_more_targets (const bool more_targets)
	Sets a flag in the connection to signal that the following connection has the same source.

bool	source_has_more_targets () const
	Returns a flag denoting whether the connection has source subsequent targets.

void	disable ()
	Disables the connection.

bool	is_disabled () const
	Returns a flag denoting if the connection is disabled.

Static Public Attributes
static constexpr ConnectionModelProperties	properties
	Properties of the connection model.

static constexpr bool	supports_flush_event = true
	Whether this connection type supports flush events.

Static Public Attributes inherited from nest::Connection< targetidentifierT >
static constexpr ConnectionModelProperties	properties = ConnectionModelProperties::NONE

static constexpr bool	supports_flush_event = false

Private Attributes
double	weight_
	Synaptic weight.

long	t_spike_previous_ = 0
	The time step when the previous spike arrived.

bool	previous_was_flush_event_ = false
	Previous event was a flush event.

long	t_previous_trigger_spike_ = 0
	The time step when the spike arrived that triggered the previous e-prop update.

double	z_bar_ = 0.0
	Low-pass filtered spiking variable.

double	e_bar_ = 0.0
	Low-pass filtered eligibility trace.

double	e_bar_reg_ = 0.0
	Low-pass filtered eligibility trace for firing rate regularization.

double	epsilon_ = 0.0
	Adaptive threshold component of the eligibility vector.

double	z_previous_buffer_ = 0.0
	Value of spiking variable one time step before t_previous_spike_.

double	gradient_ = 0.0
	Sum of gradients.

long	remaining_steps_until_cutoff_ = 0
	Remaining computation steps.

long	decay_steps_ = 0
	Decay steps for the eligibility trace.

WeightOptimizer *	optimizer_
	Optimizer.

Additional Inherited Members
Protected Member Functions inherited from nest::Connection< targetidentifierT >
void	check_connection_ (Node &dummy_target, Node &source, Node &target, const size_t receptor_type)
	This function calls check_connection() on the sender to check if the receiver accepts the event type and receptor type requested by the sender.

Protected Attributes inherited from nest::Connection< targetidentifierT >
targetidentifierT	target_

SynIdDelay	syn_id_delay_

Detailed Description

template<typename targetidentifierT>
class nest::eprop_synapse< targetidentifierT >

Class implementing a synapse model for e-prop plasticity with additional biological features.

Class implementing a synapse model for e-prop plasticity according to Bellec et al. (2020) with additional biological features described in Korcsak-Gorzo et al. (2025).

Note: Each synapse has an optimizer_ object managed through a WeightOptimizer*, pointing to an object of a specific weight optimizer type. This optimizer, drawing also on parameters in the WeightOptimizerCommonProperties accessible via the synapse models CommonProperties::optimizer_cp_ pointer, computes the weight update for the neuron. The actual optimizer type can be selected at runtime (before creating any synapses) by exchanging the optimizer_cp_ pointer. Individual optimizer objects are created by check_connection() when a synapse is actually created. It is important that the constructors of eprop_synapse do not create optimizer objects and that the destructor does not delete optimizer objects; this currently leads to bugs when using Boosts's spreadsort() due to use of the copy constructor where it should suffice to use the move constructor. Therefore, check_connection()creates the optimizer object when it is needed and specializations of Connector::~Connector() and Connector::disable_connection() delete it by calling delete_optimizer(). A disadvantage of this approach is that the default_connection in the connector model does not have an optimizer object, whence it is not possible to set default (initial) values for the per-synapse optimizer.; If we can find a way to modify our co-sorting of source and target tables in Boost's spreadsort() to only use move operations, it should be possible to create the individual optimizers in the copy constructor of eprop_synapse and to delete it in the destructor. The default_connection can then own an optimizer and default values could be set on it.

Member Typedef Documentation

◆ CommonPropertiesType

template<typename targetidentifierT >

typedef EpropSynapseCommonProperties nest::eprop_synapse< targetidentifierT >::CommonPropertiesType

Type of the common synapse properties.

◆ ConnectionBase

template<typename targetidentifierT >

typedef Connection< targetidentifierT > nest::eprop_synapse< targetidentifierT >::ConnectionBase

Type of the connection base.

Constructor & Destructor Documentation

◆ eprop_synapse() [1/3]

template<typename targetidentifierT >

nest::eprop_synapse< targetidentifierT >::eprop_synapse ( )

Default constructor.

◆ ~eprop_synapse()

template<typename targetidentifierT >

nest::eprop_synapse< targetidentifierT >::~eprop_synapse ( )

Destructor.

◆ eprop_synapse() [2/3]

template<typename targetidentifierT >

nest::eprop_synapse< targetidentifierT >::eprop_synapse ( const eprop_synapse< targetidentifierT > & es )

Parameterized copy constructor.

◆ eprop_synapse() [3/3]

template<typename targetidentifierT >

nest::eprop_synapse< targetidentifierT >::eprop_synapse ( eprop_synapse< targetidentifierT > && es )

Move constructor.

Member Function Documentation

◆ check_connection()

template<typename targetidentifierT >

void nest::eprop_synapse< targetidentifierT >::check_connection	(	Node &	s,
		Node &	t,
		size_t	receptor_type,
		const CommonPropertiesType &	cp
	)

inline

Check if the target accepts the event and receptor type requested by the sender.

Note: This sets the optimizer_ member.

References nest::WeightOptimizerCommonProperties::get_optimizer(), nest::EpropSynapseCommonProperties::optimizer_cp_, and nest::Node::register_eprop_connection().

Here is the call graph for this function:

◆ delete_optimizer()

template<typename targetidentifierT >

void nest::eprop_synapse< targetidentifierT >::delete_optimizer ( )

inline

Delete optimizer.

◆ get_delay()

template<typename targetidentifierT >

double nest::Connection< targetidentifierT >::get_delay ( ) const

inline

Return the delay of the connection in ms.

◆ get_delay_steps()

template<typename targetidentifierT >

long nest::Connection< targetidentifierT >::get_delay_steps ( ) const

inline

Return the delay of the connection in steps.

◆ get_rport()

template<typename targetidentifierT >

size_t nest::Connection< targetidentifierT >::get_rport ( ) const

inline

◆ get_status()

template<typename targetidentifierT >

void nest::eprop_synapse< targetidentifierT >::get_status ( Dictionary & d ) const

Get parameter dictionary.

References nest::names::optimizer(), nest::names::size_of(), and nest::names::weight().

Here is the call graph for this function:

◆ get_target()

template<typename targetidentifierT >

Node * nest::Connection< targetidentifierT >::get_target ( const size_t tid ) const

inline

◆ operator=() [1/2]

template<typename targetidentifierT >

eprop_synapse< targetidentifierT > & nest::eprop_synapse< targetidentifierT >::operator= ( const eprop_synapse< targetidentifierT > & es )

Assignment operator.

◆ operator=() [2/2]

template<typename targetidentifierT >

eprop_synapse< targetidentifierT > & nest::eprop_synapse< targetidentifierT >::operator= ( eprop_synapse< targetidentifierT > && es )

Move assignment operator.

◆ send()

template<typename targetidentifierT >

bool nest::eprop_synapse< targetidentifierT >::send	(	Event &	e,
		size_t	thread,
		const EpropSynapseCommonProperties &	cp
	)

Send the spike event.

◆ set_status()

template<typename targetidentifierT >

void nest::eprop_synapse< targetidentifierT >::set_status	(	const Dictionary &	d,
		ConnectorModel &	cm
	)

Update values in parameter dictionary.

References nest::names::optimizer(), and nest::names::weight().

Here is the call graph for this function:

◆ set_weight()

template<typename targetidentifierT >

void nest::eprop_synapse< targetidentifierT >::set_weight ( const double w )

inline

Set the synaptic weight to the provided value.

References nest::eprop_synapse< targetidentifierT >::weight_.

Member Data Documentation

◆ decay_steps_

template<typename targetidentifierT >

long nest::eprop_synapse< targetidentifierT >::decay_steps_ = 0

private

Decay steps for the eligibility trace.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ e_bar_

template<typename targetidentifierT >

double nest::eprop_synapse< targetidentifierT >::e_bar_ = 0.0

private

Low-pass filtered eligibility trace.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ e_bar_reg_

template<typename targetidentifierT >

double nest::eprop_synapse< targetidentifierT >::e_bar_reg_ = 0.0

private

Low-pass filtered eligibility trace for firing rate regularization.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ epsilon_

template<typename targetidentifierT >

double nest::eprop_synapse< targetidentifierT >::epsilon_ = 0.0

private

Adaptive threshold component of the eligibility vector.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ gradient_

template<typename targetidentifierT >

double nest::eprop_synapse< targetidentifierT >::gradient_ = 0.0

private

Sum of gradients.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ optimizer_

template<typename targetidentifierT >

WeightOptimizer* nest::eprop_synapse< targetidentifierT >::optimizer_

private

Optimizer.

Note: Pointer is set by check_connection() and deleted by delete_optimizer().

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ previous_was_flush_event_

template<typename targetidentifierT >

bool nest::eprop_synapse< targetidentifierT >::previous_was_flush_event_ = false

private

Previous event was a flush event.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ properties

template<typename targetidentifierT >

constexpr ConnectionModelProperties nest::eprop_synapse< targetidentifierT >::properties

staticconstexpr

Initial value:

= ConnectionModelProperties::HAS_DELAY
    | ConnectionModelProperties::IS_PRIMARY | ConnectionModelProperties::REQUIRES_EPROP_ARCHIVING
    | ConnectionModelProperties::SUPPORTS_HPC

Properties of the connection model.

Note: Does not support LBL at present because we cannot properly cast GenericModel common props in that case.

◆ remaining_steps_until_cutoff_

template<typename targetidentifierT >

long nest::eprop_synapse< targetidentifierT >::remaining_steps_until_cutoff_ = 0

private

Remaining computation steps.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ supports_flush_event

template<typename targetidentifierT >

constexpr bool nest::eprop_synapse< targetidentifierT >::supports_flush_event = true

staticconstexpr

Whether this connection type supports flush events.

◆ t_previous_trigger_spike_

template<typename targetidentifierT >

long nest::eprop_synapse< targetidentifierT >::t_previous_trigger_spike_ = 0

private

The time step when the spike arrived that triggered the previous e-prop update.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ t_spike_previous_

template<typename targetidentifierT >

long nest::eprop_synapse< targetidentifierT >::t_spike_previous_ = 0

private

The time step when the previous spike arrived.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ weight_

template<typename targetidentifierT >

double nest::eprop_synapse< targetidentifierT >::weight_

private

Synaptic weight.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=(), and nest::eprop_synapse< targetidentifierT >::set_weight().

◆ z_bar_

template<typename targetidentifierT >

double nest::eprop_synapse< targetidentifierT >::z_bar_ = 0.0

private

Low-pass filtered spiking variable.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

◆ z_previous_buffer_

template<typename targetidentifierT >

double nest::eprop_synapse< targetidentifierT >::z_previous_buffer_ = 0.0

private

Value of spiking variable one time step before t_previous_spike_.

Referenced by nest::eprop_synapse< targetidentifierT >::operator=().

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

eprop_synapse.h

	NEST main@caf0ae8

Classes

Public Types

Public Member Functions

Static Public Attributes

Private Attributes

Additional Inherited Members

Detailed Description

Member Typedef Documentation

◆ CommonPropertiesType

◆ ConnectionBase

Constructor & Destructor Documentation

◆ eprop_synapse() [1/3]

◆ ~eprop_synapse()

◆ eprop_synapse() [2/3]

◆ eprop_synapse() [3/3]

Member Function Documentation

◆ check_connection()

◆ delete_optimizer()

◆ get_delay()

◆ get_delay_steps()

◆ get_rport()

◆ get_status()

◆ get_target()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ send()

◆ set_status()

◆ set_weight()

Member Data Documentation

◆ decay_steps_

◆ e_bar_

◆ e_bar_reg_

◆ epsilon_

◆ gradient_

◆ optimizer_

◆ previous_was_flush_event_

◆ properties

◆ remaining_steps_until_cutoff_

◆ supports_flush_event

◆ t_previous_trigger_spike_

◆ t_spike_previous_

◆ weight_

◆ z_bar_

◆ z_previous_buffer_