search

cazala / synaptic

architecture-free neural network library for node.js and the browser
http://caza.la/synaptic
Other
6.92k stars 665 forks source link

Liquid State Machine Set Up #44

Open arccoxx opened 9 years ago

arccoxx commented 9 years ago

Hi,

This is a fantastic project! Good work! I was just wondering whether anyone had any thoughts on how to set up liquid state machines and specifically how they work in this module. I understand they are not the focus. Specifically, I was wondering whether there are anyways to determine pool size, connections and gates. I am finding it hard to dig up some (easily digestable) info on these networks.

Any thoughts?

Thanks!

menduz commented 9 years ago

Hi, its quite simple, as all neural networks LSM works by magic. A special kind of. You take a bunch of neurons, connect each other randomly, pick a few whos act as input, another ones will act as output. and pum! it works. Maybe you can take a look for the architect's src Liquid.ts in the brach 'typescript'. We use layers, but standalone neurons sould work too.

https://github.com/cazala/synaptic/blob/typescript/src/architect/Liquid.ts

saludos

El mar, jul 7, 2015 00:27, arccoxx notifications@github.com escribió:

Hi,

This is a fantastic project! Good work! I was just wondering whether anyone had any thoughts on how to set up liquid state machines and specifically how they work in this module. I understand they are not the focus. Specifically, I was wondering whether there are anyways to determine pool size, connections and gates. I am finding it hard to dig up some (easily digestable) info on these networks.

Any thoughts?

Thanks!

— Reply to this email directly or view it on GitHub https://github.com/cazala/synaptic/issues/44.

arccoxx commented 9 years ago

Thanks for the speedy response!

I was concerned more on how to decide the number of pool neurons, connections and gates when building the network. How do I decide on these parameters? Also, is the network self optimizing?

thanks!

arccoxx commented 9 years ago

Oh hold on I see what the confusion is. I'm interested in the actual design of liquid state networks not how I create them with the module. Specifically based on input output how do I determine the number of neuron gates and connections?

Hope that clears things up.

cazala commented 9 years ago

That's the tricky part, determining the topology of the network (specially with liquid state machines) sometimes can be a matter of trial and error, some approaches use genetic algorithms to find most efficient configurations: create a set of networks with random amounts of neurons/connections, test them solving your problem, keep the ones that do the best job, create a new set of networks using similar configurations (crossover) and repeat the process till you find the topology that suits your problem.

mamcn commented 6 years ago

Sorry! brother @menduz which programming language did you used for LSM code

menduz commented 6 years ago

https://typescriptlang.org