spark.nn.components.learning_rules

spark.nn.components.learning_rules#

Submodules#

Classes#

`LearningRule`	Abstract learning rule model.
`LearningRuleConfig`	Abstract learning rule configuration class.
`LearningRuleOutput`	Generic learning rule model output spec.
`ZenkeRule`	Zenke plasticy rule model. This model is an extension of the classic Hebbian Rule.
`ZenkeRuleConfig`	ZenkeRule configuration class.
`HebbianRule`	Hebbian plasticy rule model.
`HebbianRuleConfig`	HebbianRule configuration class.
`OjaRule`	Oja's plasticy rule model.
`OjaRuleConfig`	Abstract learning rule configuration class.
`QuadrupletRule`	Quadruplet plasticy rule model.
`QuadrupletRuleConfig`	QuadrupletRule configuration class.
`QuadrupletRuleTensor`	Quadruplet plasticy rule model (tensor).
`QuadrupletRuleTensorConfig`	QuadrupletRuleTensor configuration class.
`ThreeFactorHebbianRule`	Three-factor Hebbian plasticy rule model.
`ThreeFactorHebbianRuleConfig`	ThreeFactorHebbianRule configuration class.

Package Contents#

class spark.nn.components.learning_rules.LearningRule(config=None, **kwargs)[source]#

Bases: spark.nn.components.base.Component, Generic[ConfigT]

Abstract learning rule model.

Parameters:: config (ConfigT | None)

class spark.nn.components.learning_rules.LearningRuleConfig(__skip_validation__=False, **kwargs)[source]#

Bases: spark.nn.components.base.ComponentConfig

Abstract learning rule configuration class.

Parameters:: __skip_validation__ (bool)

class spark.nn.components.learning_rules.LearningRuleOutput[source]#

Bases: TypedDict

Generic learning rule model output spec.

Initialize self. See help(type(self)) for accurate signature.

kernel: spark.core.payloads.FloatArray[source]#

class spark.nn.components.learning_rules.ZenkeRule(config=None, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRule

Zenke plasticy rule model. This model is an extension of the classic Hebbian Rule.

Init:: pre_tau: float | jax.Array post_tau: float | jax.Array post_slow_tau: float | jax.Array target_tau: float | jax.Array a: float | jax.Array b: float | jax.Array c: float | jax.Array d: float | jax.Array P: float | jax.Array eta: float | jax.Array
Input:: pre_spikes: SpikeArray post_spikes: SpikeArray kernel: FloatArray
Output:: kernel: FloatArray

Parameters:: config (ZenkeRuleConfig | None)

config: ZenkeRuleConfig[source]#

build(input_specs)[source]#

Build method.

Parameters:: input_specs (dict[str, spark.core.specs.PortSpecs])

reset()[source]#

Resets component state.

Return type:: None

__call__(pre_spikes, post_spikes, kernel)[source]#

Computes and returns the next kernel update.

Parameters:

pre_spikes (spark.core.payloads.SpikeArray)
post_spikes (spark.core.payloads.SpikeArray)
kernel (spark.core.payloads.FloatArray)

Return type:

spark.nn.components.learning_rules.base.LearningRuleOutput

class spark.nn.components.learning_rules.ZenkeRuleConfig(__skip_validation__=False, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRuleConfig

ZenkeRule configuration class.

Parameters:: __skip_validation__ (bool)

pre_tau: float | jax.Array | spark.nn.initializers.base.Initializer[source]#

post_tau: float | jax.Array | spark.nn.initializers.base.Initializer[source]#

post_slow_tau: float | jax.Array | spark.nn.initializers.base.Initializer[source]#

target_tau: float | jax.Array | spark.nn.initializers.base.Initializer[source]#

a: float | jax.Array[source]#

b: float | jax.Array[source]#

c: float | jax.Array[source]#

d: float | jax.Array[source]#

p: float | jax.Array[source]#

eta: float | jax.Array[source]#

class spark.nn.components.learning_rules.HebbianRule(config=None, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRule

Hebbian plasticy rule model.

Init:: pre_tau: float | jax.Array post_tau: float | jax.Array gamma: float | jax.Array
Input:: pre_spikes: SpikeArray post_spikes: SpikeArray kernel: FloatArray
Output:: kernel: FloatArray

Parameters:: config (HebbianRuleConfig | None)

config: HebbianRuleConfig[source]#

build(input_specs)[source]#

Build method.

Parameters:: input_specs (dict[str, spark.core.specs.PortSpecs])

reset()[source]#

Resets component state.

Return type:: None

__call__(pre_spikes, post_spikes, kernel)[source]#

Computes and returns the next kernel update.

Parameters:

pre_spikes (spark.core.payloads.SpikeArray)
post_spikes (spark.core.payloads.SpikeArray)
kernel (spark.core.payloads.FloatArray)

Return type:

spark.nn.components.learning_rules.base.LearningRuleOutput

class spark.nn.components.learning_rules.HebbianRuleConfig(__skip_validation__=False, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRuleConfig

HebbianRule configuration class.

Parameters:: __skip_validation__ (bool)

pre_tau: float | jax.Array | spark.nn.initializers.base.Initializer[source]#

post_tau: float | jax.Array | spark.nn.initializers.base.Initializer[source]#

eta: float[source]#

class spark.nn.components.learning_rules.OjaRule(config=None, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRule

Oja’s plasticy rule model.

Init:: pre_tau: float | jax.Array post_tau: float | jax.Array gamma: float | jax.Array
Input:: pre_spikes: SpikeArray post_spikes: SpikeArray kernel: FloatArray
Output:: kernel: FloatArray

Parameters:: config (OjaRuleConfig | None)

config: OjaRuleConfig[source]#

build(input_specs)[source]#

Build method.

Parameters:: input_specs (dict[str, spark.core.specs.PortSpecs])

reset()[source]#

Resets component state.

Return type:: None

__call__(pre_spikes, post_spikes, kernel)[source]#

Computes and returns the next kernel update.

Parameters:

pre_spikes (spark.core.payloads.SpikeArray)
post_spikes (spark.core.payloads.SpikeArray)
kernel (spark.core.payloads.FloatArray)

Return type:

spark.nn.components.learning_rules.base.LearningRuleOutput

class spark.nn.components.learning_rules.OjaRuleConfig(__skip_validation__=False, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRuleConfig

Abstract learning rule configuration class.

Parameters:: __skip_validation__ (bool)

post_tau: float | jax.Array | spark.nn.initializers.base.Initializer[source]#

eta: float[source]#

class spark.nn.components.learning_rules.QuadrupletRule(config=None, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRule

Quadruplet plasticy rule model.

Init:: pre_tau: float | jax.Array post_tau: float | jax.Array q_alpha: float | jax.Array q_beta: float | jax.Array q_gamma: float | jax.Array q_delta: float | jax.Array eta: float | jax.Array
Input:: modulation: FloatArray pre_spikes: SpikeArray post_spikes: SpikeArray kernel: FloatArray
Output:: kernel: FloatArray

Parameters:: config (QuadrupletRuleConfig | None)

config: QuadrupletRuleConfig[source]#

build(input_specs)[source]#

Build method.

Parameters:: input_specs (dict[str, spark.core.specs.PortSpecs])

reset()[source]#

Resets component state.

Return type:: None

__call__(modulation, pre_spikes, post_spikes, kernel)[source]#

Computes and returns the next kernel update.

Parameters:

modulation (spark.core.payloads.FloatArray)
pre_spikes (spark.core.payloads.SpikeArray)
post_spikes (spark.core.payloads.SpikeArray)
kernel (spark.core.payloads.FloatArray)

Return type:

spark.nn.components.learning_rules.base.LearningRuleOutput

class spark.nn.components.learning_rules.QuadrupletRuleConfig(__skip_validation__=False, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRuleConfig

QuadrupletRule configuration class.

Parameters:: __skip_validation__ (bool)

pre_tau: float | jax.Array[source]#

post_tau: float | jax.Array[source]#

q_alpha: float | jax.Array[source]#

q_beta: float | jax.Array[source]#

q_gamma: float | jax.Array[source]#

q_delta: float | jax.Array[source]#

gamma: float | jax.Array[source]#

class spark.nn.components.learning_rules.QuadrupletRuleTensor(config=None, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRule

Quadruplet plasticy rule model (tensor).

Init:: pre_tau: float | jax.Array post_tau: float | jax.Array eta: float | jax.Array
Input:: modulation: FloatArray pre_spikes: SpikeArray post_spikes: SpikeArray kernel: FloatArray
Output:: kernel: FloatArray

Parameters:: config (QuadrupletRuleTensorConfig | None)

config: QuadrupletRuleTensorConfig[source]#

build(input_specs)[source]#

Build method.

Parameters:: input_specs (dict[str, spark.core.specs.PortSpecs])

reset()[source]#

Resets component state.

Return type:: None

__call__(modulation, pre_spikes, post_spikes, kernel)[source]#

Computes and returns the next kernel update.

Parameters:

modulation (spark.core.payloads.FloatArray)
pre_spikes (spark.core.payloads.SpikeArray)
post_spikes (spark.core.payloads.SpikeArray)
kernel (spark.core.payloads.FloatArray)

Return type:

spark.nn.components.learning_rules.base.LearningRuleOutput

class spark.nn.components.learning_rules.QuadrupletRuleTensorConfig(__skip_validation__=False, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRuleConfig

QuadrupletRuleTensor configuration class.

Parameters:: __skip_validation__ (bool)

pre_tau: tuple[float, float, float, float][source]#

post_tau: tuple[float, float, float, float][source]#

q_alpha: tuple[float, float, float, float][source]#

q_beta: tuple[float, float, float, float][source]#

q_gamma: tuple[float, float, float, float][source]#

q_delta: tuple[float, float, float, float][source]#

max_clip: tuple[float, float, float, float][source]#

eta: float[source]#

class spark.nn.components.learning_rules.ThreeFactorHebbianRule(config=None, **kwargs)[source]#

Bases: spark.nn.components.learning_rules.base.LearningRule

Three-factor Hebbian plasticy rule model.

Init:: pre_tau: float | jax.Array post_tau: float | jax.Array gamma: float | jax.Array
Input:: pre_spikes: SpikeArray post_spikes: SpikeArray kernel: FloatArray
Output:: kernel: FloatArray