To perform a simulation in N2S3 we need to introduce some stimuli into the network. This stimuli will help us understanding how our simulated network learns and reacts to events. In N2S3 these stimuli can come from existing file datasets such as MNIST, or we can generate it in memory.

The section belows explain the current internal design of input stimuli in N2S3. This may give an idea on how to design new input stimuli

N2S3 implements a separate actor for taking care of the input stimuli: the InputLayerActor. Internally, the behaviour of an InputLayerActor is defined in the class InputLayer. The actor structure is as follows:

An important point in this design is that the InputLayer behaves as a neuron group. That is,

• It virtually contains as many (input) neurons as possible different locations of spiking there is in the input data. For example, to make a network that accepts images of 10×10 pixels, we need 100 input neurons.
• It can be connected to other neuron groups using the already existing connection policies.

Besides this, an InputLayer is configured with an input stream. It is the input layer that reads from the input stream, generate spikes and forward them to the next layer following the configured connections.

The input actor will read the input from an input stream. This input stream implements basically a decorator pattern as follows, where a basic entry is decorated by a more complex decorator containing a converter: