Enhancing an existing learning rule

After we shed some light on the coding of information into spike trains, we went one step further: how do these spiking neural networks have to be trained to solve our engineering problems? Several training techniques are possible: local Hebbian-based learning, global black-box optimisation like genetic algorithms, and supervised learning rules. Supervised learning rules are the most interesting in light of solving engineering problems.
Currently only one supervised learning rule exists: SpikeProp. It was presented by S. Bohte in 2000. It is a learning rule that is applicable to spiking neural networks that use time-to-first spike coding. This type of coding only looks at the first spike, then the system is completely reset.
In "Extending SpikeProp" presented at IJCNN2004, I presented several new training rules to the SpikeProp algorithm. Due to this times less weights are needed to learn the binairy XOR function.