This repo the implementation of SpiKL-IP in the resevoir on the LSM network.
The code can be run in parallel to crossvalidate Speech data TI46.
The papar Information-Theoretic Intrinsic Plasticity for Online Unsupervised Learning in Spiking Neural Networks is published by the Frontiers in Neuroscience.
Contact [email protected] if you have any questions or concerns.
- neuronGroup: Use this to create a group of neurons
neurongroup [name] [num_neurons] [excitatory/inhibitory] vmem
- column: Use this to create the reservoir
column [name] [x] [y] [z]
x, y, z are three dimensions of the reservoir
- lsmsynapse: Specify the connections
lsmsynapse [input_neuron_group] [output_neuron_group] [input_connect] [output_connect] [weight_max] [mode] [fixed/learning]
input_connect, output_connect:
Select $input_connect neurons to randomly connect to $output_connect neurons.
Notice that the input_connect can only be 1 or -1
If input_connect == -1 and output_connect == -1, that means the fully connectivity
mode = 0, 1, 2
0: initialize all weights as zero
1: initialize all weights as a fixed number (weight_max)
2: initialize the weights randomly in (-1, 1)
fixed/learning : tell you whether the synapse can be learned or not.
$ make -j
$ ./NeuromorphicSimIn the def.h file, most of neuron parameters can be changed. Users can enable SpiKL_IP to test network with IP, and disable that define to test network without IP.
The spikes generated in the input layer and reservoir will be written in the files in fold spikes. Each file has two columns. The first one represents index of neuron and the second one represents spike time.