synapse_traits library

The synapse traits library is a set of synapse attributes. You can use the synapses provided in the library, for example, when creating projections.

Currently, the library supports the property sets for the following synapses:

• Synapse based on the DeltaSynapse model (hereinafter also referred to as delta synapse).

  After receiving spikes from a presynaptic population, the synapses generate a message containing the synaptic impact values. The synapse projection sends the synaptic impact to the population to change the membrane potential of the postsynaptic neuron. In this case, the message is sent by the projection without a time delay.

  

  A container containing a set of synapses of the same type connecting neurons of the presynaptic and postsynaptic populations.

  

  A value generated by a synapse for changing the values of the neuron attributes.

  

  A container containing a set of neurons and their function based on the neuron type.

  

  Short-term action potential generated by a neuron in a population as a result of the neuron attribute changes.

  The sent synaptic impact is determined by the weight of the delta synapse. You can specify the synapse weight in the attributes of its instance.

  

  The value of the signal that the synapse sends to the neurons of the associated population.

• Synapse based on the AdditiveSTDPDeltaSynapse model.

  A synapse based on the AdditiveSTDPDeltaSynapse model is a delta synapse whose synaptic plasticity depends on the time of spike receipt. If a spike from the presynaptic population is received by the projection before the postsynaptic neuron generates the presynaptic spike, the synapse weight value increases. If a presynaptic spike is received by the synapse after the postsynaptic neuron generates a spike, the weight of the synapse decreases.

  The synapse weight determines the value of the synaptic impact sent to the postsynaptic population.

  The change in synapse weight is calculated using the formula shown below, where:

  • j corresponds to a presynaptic neuron.
  • i corresponds to a postsynaptic neuron.
  • W corresponds to the synaptic plasticity function.
  • n corresponds to the sequential number of the presynaptic neuron.
  • f corresponds to the sequential number of the postsynaptic neuron.
  • corresponds to the time when the synapse receives spikes from a presynaptic neuron.
  • corresponds to the time when the postsynaptic neuron generates the spike.
  • corresponds to the value of the change in synapse weight after a spike is received from a presynaptic neuron.

  

  Formula for calculating the change in synapse weight based on the AdditiveSTDPDeltaSynapse model

• Synapse based on the SynapticResourceSTDPDeltaSynapse model.

  A synapse based on the SynapticResourceSTDPDeltaSynapse model is a delta synapse whose synaptic plasticity depends on the synaptic resource. The value of the synaptic resource depends on the synaptic plasticity:

  • Hebbian plasticity.

    The value of a synaptic resource depends on the period of the dense sequence of spikes (hereinafter also referred to as DSS). DSS is a sequence of unforced spikes generated by a neuron, where all adjacent spikes are separated by the time intervals by no more than ISImax, where ISImax is the maximum allowed distance between spikes in the DSS. The DSS period is the time interval between the moment when the Hebbian plasticity starts before the first spike in the DSS and one of the following moments: when the maximum allowable distance ISImax is reached after the last spike in the DSS, when a forced activation occurs, or when the spike arrives at the dopamine synapse.

    When at least one spike is received within the DSS period, the synaptic resource changes by the value of the current Hebbian plasticity. The Hebbian plasticity depends on the neuron stability variable and is calculated using the formula shown below, where:

    • corresponds to Hebbian plasticity.
    • s corresponds to the neuron stability variable. At the time of the first spike within the DSS period, the stability value changes by a constant. If the stability is negative, it does not change downward.

  

  Formula for calculating Hebbian plasticity

  • Unconditional plasticity.

    In the case of a nonzero unconditioned plasticity, the value of the synaptic resource decreases by the value of plasticity when a spike is received. The unconditional plasticity depends on the neuron stability variable and is calculated using the formula shown below, where:

    • corresponds to unconditional plasticity.
    • s corresponds to the neuron stability variable. At the time of the first spike within the DSS period, the stability value changes by a constant. If the stability is negative, it does not change downward.

  

  Formula for calculating unconditional plasticity

  • Dopamine plasticity.

    The dopamine plasticity is applicable when the sum of all dopamine synapses that receive a spike at the current step of the neural network execution is not zero. In the case of dopamine plasticity, the value of the synaptic resource changes based on the formula below, where:

    • D corresponds to the sum of dopamine synapses.
    • s corresponds to the neuron stability variable.

  

  Formula for calculating synaptic resource in case of dopamine plasticity

  The sum of dopamine synapses affects the neuron stability variable. If at the time of the forced activation, the dopamine punishment is received (the sum of dopamine synapses is less than zero), the value of the neuron stability is decreased by the rD value, where r is a constant and D is dopamine punishment.

  If at the time of the forced activation, the dopamine reward is received (the sum of dopamine synapses is greater than zero), the value of the neuron stability is increased by the rD value, where r is a constant and D is dopamine reward. Otherwise, the stability value changes to the value shown below, where:

  • D corresponds to the sum of dopamine synapses.
  • ISImax corresponds to the maximum acceptable distance between spikes in the DSS.
  • corresponds to the time since the first spike in the last DSS.

  

  Formula for calculating the neuron stability change in case of the dopamine reward

  The synapse weight determines the value of the synaptic impact sent to the postsynaptic population and is associated with the value of the synaptic resource by the formula shown below, where:

• w corresponds to the synapse weight.
• W corresponds to a synaptic resource.
• corresponds to a random minimum weight.
• corresponds to a random maximum weight.

  

  Formula for calculating the weight change depending on the synaptic resource

For more details on the properties of synapses, see API documentation for Kaspersky Neuromorphic Platform. If needed, you can implement and use your own synapse models when working with source code.

Page top