some aspects to start with are the following questions:
- what is your desired firing rate? --> set the refractory time (in LIF) to set the maximum firing rate and leakage current/conductances accordingly; many other parameters determine the range of firing rate in spiking neurons (including the connection weights)
For a minimum (base/resting) firing rate, you may need to add some input current / background noise which drives the cells constantly / randomly.
- does your model require some specific temporal firing patterns (if so, you may want to have a look at "Which Model to Use for Cortical Spiking Neurons?" by Izhikevich 2004
- how many neurons do you need and which simulator do you want to use? This is important for the computational costs and complexity. You may want to have a look at: Brette et al. 2007 "Simulation of networks of spiking neurons: A review of tools and strategies"
I'm sure this does not answer your question completely, but since I'm not an expert in implementing RBMs these should merely be some points to start with.
Thank you dear Bernhard. your answer was very useful. In fact I'm not expert in RBM and also in spiking models. I have just implemented a RBM model in MATLAB and now I'm trying to make some changes and make the RBM (which has a good result in some jobs) more close to biological model (which has some benefits like less power consumption ). I saw some papers and thesis (By Peter O'coner, Dan Neil and ...) which are very close to what I want, but I need to know more. Because of the verity and huge amount of topics, really I don't know what I have to do and what I have to read?????
some publications that you might find interesting are:
1) Brown, Andrew D. "Spiking boltzmann machines." In Advances in Neural Information Processing Systems 12: Proceedings of the 1999 Conference, vol. 12, p. 122. MIT Press, 2000.
2) Buesing, Lars, Johannes Bill, Bernhard Nessler, and Wolfgang Maass. "Neural dynamics as sampling: a model for stochastic computation in recurrent networks of spiking neurons." PLoS Comput. Biol 7, no. 11 (2011): e1002211.
This paper basically follows the idea that probabilities are represented as spikes (which is somewhat accepted and not at all new), but since they use Boltzmann distributions you might be able to link it to RBMs as well (as I said, I'm not an expert in RBMs). However, their spiking neurons are not at all biologically plausible - but maybe that's not important for you?
-->The question remains: what are the requirements that you have for using spikes? Answering this may help you find a suitable spiking neuron model. And: do they need to be biologically plausible and why? Maybe a spike-response model would be also suitable? http://www.scholarpedia.org/article/Spike-response_model