Adding a new neuron type

Currently, the Kaspersky Neuromorphic Platform supports the property sets and population typification based on the neuron models AltAILIF, BLIFATNeuron, and SynapticResourceSTDPBLIFATNeuron. If required, you can add a new neuron type.

You can use this instruction when working with the platform source code.

To add a new neuron type:

1. Go from the working directory where the platform source code archive was unpacked to neuron-trait-library/include/knp/neuron-traits/, and create a header file for the new neuron type there.
2. In the created header file, define the neuron type structure and its template properties (for example, default_values, neuron_parameters).

   An example of a BLIFAT neuron definition:

   neuron-trait-library/include/knp/neuron-traits/blifat.h

   #include "type_traits.h"

   namespace knp::neuron_traits

   {

   struct BLIFATNeuron;

   template <>

   struct default_values<BLIFATNeuron>

   {

   // Specifies the number of steps in the neural network since the last spike step by default.

   constexpr static std::size_t n_time_steps_since_last_firing_ = std::numeric_limits<std::size_t>::infinity();

   // Specifies the default value to which the membrane potential tends for

   // conduction-based inhibitory synapses.

   constexpr static double reverse_inhibitory_potential = -0.3;

   // Specifies the default value to which the membrane potential tends for

   // current-based inhibitory synapses.

   constexpr static double min_potential = -1.0e9;

   // ...

   };

   template <>

   struct neuron_parameters<BLIFATNeuron>

   {

   // Specifies the number of steps in the neural network since the last spike step.

   std::size_t n_time_steps_since_last_firing_ = default_values<BLIFATNeuron>::n_time_steps_since_last_firing_;

   // Specifies the threshold value of the membrane potential of the neuron.

   double activation_threshold_ = default_values<BLIFATNeuron>::activation_threshold_;

   // Specifies a dynamic threshold value of the membrane potential for the neuron to reach

   // before it generates a spike.

   double dynamic_threshold_ = default_values<BLIFATNeuron>::dynamic_threshold_;

   // ...

   };

   } // namespace knp::neuron_traits

3. Add the created neuron type to the list of neurons defined in the neuron-trait-library/include/knp/neuron-traits/all_traits.h header file.

   An example of adding a BLIFAT neuron to the list of neurons:

   neuron-trait-library/include/knp/neuron-traits/all_traits.h

   #include "blifat.h"

   namespace knp::neuron_traits

   {

   // A comma-separated list of neuron types.

   #define ALL_NEURONS BLIFATNeuron

   // ...

   } // namespace knp::neuron_traits

4. Implement the overloading startup method in the desired backends.

   For example, for the CPU backend, implement the overloading method for starting the calculation of populations.

Page top