A Novel Self-Organised Learning Model with Temporal Coding for Spiking Neural Networks

In this paper, a novel self-organised learning model with temporal coding is proposed for a network of spiking neurons which encode information through the timing of action potentials. The development of this learning model is based on recent findings in biological neural systems. The Hebbian-type learning equation for the proposed model utilises the time difference between the input and output spikes. The proposed spiking neural network learning model was tested on two sets of benchmark data. Clusters were formed in the output space based on the position of the output neurons and their firing time. The results show that networks trained using action potential timings are capable of learning complex tasks.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

NOTE: TO READ THIS AND OTHER IPROMS 2006 PAPERS, PLEASE REGISTER FOR THE CONFERENCE.

REGISTRATION IS FREE.

CLICK here TO REGISTER.

Submitted by charles on Tue, 27/06/2006 - 9:40am.
login or register to post comments |

topology preservation

Submitted by gonzalo on Thu, 06/07/2006 - 4:49pm.

Hi Charles, excellent work!!! This is the first time I see SOM using SNN. I have the following question: Kohonen’s SOM apart from generating a topographical mapping of the input space also tries to achieve good topology preservation. Have you run any test to measure how good the topology preservation is? Does it compare with Kohonen’s SOM?

Cheers Mate

Gonzalo

Regarding the topology preservation

Submitted by charles on Fri, 07/07/2006 - 8:08am.

Hello Gonzalo,

A top logical question, thank you.
I didnt do any specific test to measure the preservation of the topology of the input space. But this model of spiking neural networks are much realistic as biological neural networks than the sigmoidal networks, and the learning model is similar to Kohonen's SOM, it could be expected to preserve the topology of the input space. And one more thing, a phenomenon observed in learning; when the bench mark data sets were clusterd, the classes which are linearly seprable were found to be represented with different groups of neurons. But the linearly non separable classes were found to be represented with a same set of neurons and separated with their firing times. Hope this could give you a good idea.

cheers,
Charles.

Novel approach for learning in SNN

Submitted by charles on Wed, 12/07/2006 - 9:33am.

Hi,

The Spiking neural network is claimed to be the third generation of neural network and has much more advantages over the existing neural network models. But until now not much of its potential has not been utilised. The main reason could be that the learning modles proposed are based on the exisitng techniques for sigmoidal neural networks. There is a wide belief that new approaches not based on the existing learning models could be a key.

I am expecting your comments regarding this.

regards,
Charles.

Multiple neural spike train data analysis

Submitted by yiw on Mon, 24/07/2006 - 2:56pm.

hi:

I just read tha paper Multiple neural spike train data analysis
http://www.nature.com/neuro/journal/v7/n5/full/nn1228.html

is your analysis based single spiker or multi spike?

best regards

yi wang

Hello Yi, Thanks for the

Submitted by charles on Tue, 25/07/2006 - 4:47pm.

Hello Yi,

Thanks for the query.

My model is based on single spikes. The multi spike models could be even more powerfull but decoding the spike trains would be complex.

cheers,
Charles.

A Novel Self-Organised Learning Model with Temporal Coding for Spiking Neural Networks

Comment viewing options

User login