Liquid State Machines Training

[neurogeek953]

Dear all,

I am currently looking for information on how do you train Liquid State Machines.
I would like to know more about the algorithm you are using to train your spiking neural networks and find documentation on this if it is available.

Would it be possible for you to give me the information?

Thank you,
neurogeek953

[urbain]

Hi neurogeek953,

Due to their architecture, Liquid State Machines allows simple supervised training algorithms to be used:

• The simplest way is clearly to first generate all the reservoir states and teaching signals and the use linear regression (you can find this in any good algebra books) to get the output weights
• If you need to train your reservoir online (i.e. to perform a training action at every time step), you can use Recurrent Least Square algorithm or FORCE learning. We use the latter for creating Central Pattern Generators for quadruped locomotion

I hope this answer your question!
Best,

Gabriel

[neurogeek953]

Thanks urbain!

This was exactly what I was looking for!

Best,
neurogeek953

[neurogeek953]

Hello Urbain,

By any chance do you know where the connectivity map between the neural-nets nodes in these algorithms is stored?

Best,
neurogeek953