ml_genn.neurons package

Neuron models describe the dynamics and spiking behaviour of a Population.

class ml_genn.neurons.AdaptiveLeakyIntegrateFire(v_thresh=1.0, v_reset=0.0, v=0.0, a=0.0, beta=0.0174, tau_mem=20.0, tau_refrac=None, tau_adapt=2000.0, relative_reset=True, integrate_during_refrac=True, readout=None)

Bases: Neuron

A leaky-integrate and fire neuron with an adaptive firing threshold as described by [Bellec2018].

Parameters:

  • v_thresh (InitValue) – Membrane voltage firing threshold

  • v_reset (InitValue) – After a spike is emitted, this value is subtracted from the membrane voltage v if relative_reset is True. Otherwise, if relative_reset is False, the membrane voltage is set to this value.

  • v (InitValue) – Initial value of membrane voltage

  • a (InitValue) – Initial value of adaptation

  • beta (InitValue) – Strength of adaptation

  • tau_mem (InitValue) – Time constant of membrane voltage [ms]

  • tau_refrac (InitValue) – Duration of refractory period [ms]

  • tau_adapt (InitValue) – Time constant of adaptation [ms]

  • relative_reset (bool) – How is v reset after a spike?

  • integrate_during_refrac (bool) – Should v continue to integrate inputs during refractory period?

  • readout – Type of readout to attach to this neuron’s output variable

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.BinarySpikeInput(signed_spikes=False, input_frames=1, input_frame_timesteps=1)

Bases: Neuron, InputBase

Input neuron which simply emits a spike if the input is greater than zero. Optionally, it can also emit a ‘negative’ spike if input is less than zero.

Parameters:

  • signed_spikes – Should negative spikes be emitted if input is less than zero?

  • input_frames – How many frames does each input have?

  • input_frame_timesteps – How many timesteps should each frame of input be presented for?

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.FewSpikeRelu(k=10, alpha=25, readout=None)

Bases: Neuron

A few-spike neuron to encode a ReLU ANN activation as described by [Stockl2021].

Should typically be created by converting an ANN to an SNN using ml_genn_tf.converters.FewSpike.

Parameters:

  • k (int) – Number of timesteps to encode activation over.

  • alpha (float) – Scaling factor to apply to activations.

  • readout – Type of readout to attach to this neuron’s output variable

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

pipelined = True
class ml_genn.neurons.FewSpikeReluInput(k=10, alpha=25, signed_input=False)

Bases: Neuron, InputBase

A few-spike neuron to encode inputs using a ReLU activation as described by [Stockl2021].

Should typically be created by converting an ANN to an SNN using ml_genn_tf.converters.FewSpike.

Parameters:

  • k (int) – Number of timesteps to encode activation over.

  • alpha (float) – Scaling factor to apply to activations.

  • signed_input – Are inputs expected to be both positive and negative?

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.Input

Bases: object

Base class for all types of input neuron

abstract set_input(genn_pop, batch_size, shape, input)

Copy provided data to GPU.

Parameters:

  • genn_pop – GeNN NeuronGroup object population has been compiled into

  • batch_size (int) – Batch size of compiled network

  • shape – Shape of input population

  • input – Input data

class ml_genn.neurons.IntegrateFire(v_thresh=1.0, v_reset=0.0, v=0.0, readout=None, **kwargs)

Bases: Neuron

An integrate and fire neuron.

Parameters:

  • v_thresh (InitValue) – Membrane voltage firing threshold

  • v_reset (InitValue) – After a spike is emitted, the membrane voltage is set to this value.

  • v (InitValue) – Initial value of membrane voltage

  • readout – Type of readout to attach to this neuron’s output variable

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.IntegrateFireInput(v_thresh=1.0, v_reset=0.0, v=0.0, input_frames=1, input_frame_timesteps=1)

Bases: IntegrateFire, InputBase

An integrate and fire input neuron.

Parameters:

  • v_thresh (InitValue) – Membrane voltage firing threshold

  • v_reset (InitValue) – After a spike is emitted, the membrane voltage is set to this value.

  • v (InitValue) – Initial value of membrane voltage

  • input_frames – How many frames does each input have?

  • input_frame_timesteps – How many timesteps should each frame of input be presented for?

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.LeakyIntegrate(v=0.0, bias=0.0, tau_mem=20.0, scale_i=False, readout=None)

Bases: Neuron

A leaky-integrator, typically used as an output neuron

Parameters:

  • v (InitValue) – Initial value of membrane voltage

  • bias (InitValue) – Initial value of bias curremt

  • tau_mem (InitValue) – Time constant of membrane voltage [ms]

  • scale_i (bool) – Should input current x be scaled by \(1-e^\frac{-dt}{\tau_\text{mem}}\)?

  • readout – Type of readout to attach to this neuron’s output variable

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.LeakyIntegrateFire(v_thresh=1.0, v_reset=0.0, v=0.0, tau_mem=20.0, tau_refrac=None, relative_reset=True, integrate_during_refrac=True, scale_i=False, readout=None, **kwargs)

Bases: Neuron

A leaky-integrate and fire neuron.

Parameters:

  • v_thresh (InitValue) – Membrane voltage firing threshold

  • v_reset (InitValue) – After a spike is emitted, this value is subtracted from the membrane voltage v if relative_reset is True. Otherwise, if relative_reset is False, the membrane voltage is set to this value.

  • v (InitValue) – Initial value of membrane voltage

  • tau_mem (InitValue) – Time constant of membrane voltage [ms]

  • tau_refrac (InitValue) – Duration of refractory period [ms]

  • relative_reset (bool) – How is v reset after a spike?

  • integrate_during_refrac (bool) – Should v continue to integrate inputs during refractory period?

  • readout – Type of readout to attach to this neuron’s output variable

  • scale_i (bool)

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.LeakyIntegrateFireInput(v_thresh=1.0, v_reset=0.0, v=0.0, tau_mem=20.0, tau_refrac=None, relative_reset=True, integrate_during_refrac=True, scale_i=False, input_frames=1, input_frame_timesteps=1)

Bases: LeakyIntegrateFire, InputBase

A leaky-integrate and fire input neuron.

Parameters:

  • v_thresh (InitValue) – Membrane voltage firing threshold

  • v_reset (InitValue) – After a spike is emitted, this value is subtracted from the membrane voltage v if relative_reset is True. Otherwise, if relative_reset is False, the membrane voltage is set to this value.

  • v (InitValue) – Initial value of membrane voltage

  • tau_mem (InitValue) – Time constant of membrane voltage [ms]

  • tau_refrac (InitValue) – Duration of refractory period [ms]

  • relative_reset (bool) – How is v reset after a spike?

  • integrate_during_refrac (bool) – Should v continue to integrate inputs during refractory period?

  • input_frames – How many frames does each input have?

  • input_frame_timesteps – How many timesteps should each frame of input be presented for?

  • scale_i (bool)

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.Neuron(readout=None, **kwargs)

Bases: ABC

Base class for all neuron models

Variables:

readout – Type of readout to attach to this neuron’s output variable

Parameters:

readout (Optional[Readout])

abstract get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

get_readout(genn_pop, batch_size, shape)

Use readout object associated with neuon to read output from PyGeNN neuron group into numpy array

Parameters:

  • genn_pop – PyGeNN neuron group

  • batch_size (int) – Batch size of compiled model

  • shape – Shape of population

Return type:

ndarray

property readout

Optional object which can be used to provide a readout from neuron

Can be specified as either a Readout object or, for built in readout models whose constructors require no arguments, a string e.g. “spike_count”

class ml_genn.neurons.PoissonInput(signed_spikes=False, input_frames=1, input_frame_timesteps=1)

Bases: Neuron, InputBase

Input neuron which generates spikes using a Poisson process whose rate is proportional to the magnitude of the input.

Parameters:

  • signed_spikes – Should negative spikes be emitted if input is less than zero?

  • input_frames – How many frames does each input have?

  • input_frame_timesteps – How many timesteps should each frame of input be presented for?

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

class ml_genn.neurons.SpikeInput(max_spikes=1000000)

Bases: Neuron, Input

An input neuron which emits spike trains provided as ml_genn.utils.data.PreprocessedSpikes objects

Parameters:

max_spikes – Maximum total number of spikes this population can emit for any example. This needs to include batch size i.e. if batch size is 32 and there are 100 neurons in the population, each of which emits a maximum of one spike per example, \(\text{max_spikes} = 32 * 100 * 1 = 3200\)

get_model(population, dt, batch_size)

Gets PyGeNN implementation of neuron model

Parameters:

  • population (Population) – Population this neuron is to be attached to

  • dt (float) – Timestep of simulation (in ms)

  • batch_size (int) – Batch size of the model

Return type:

NeuronModel

set_input(genn_pop, batch_size, shape, input)

Copy provided data to GPU.

Parameters:

  • genn_pop – GeNN NeuronGroup object population has been compiled into

  • batch_size (int) – Batch size of compiled network

  • shape – Shape of input population

  • input (PreprocessedSpikes | Sequence[PreprocessedSpikes]) – Input data