[PD] pm mapping

[cyrille henry]

hello,

Georg Holzmann a écrit :
> Hallo!
> 
>> so, we need to imagine a n dimentional input parametter, "link" to m 
>> dimentional output.
>> this "active mapping space" implementation should be quite easy once 
>> i've got time + iem matrix.
> 
> I didn't get much from the rest of the thread, so sorry if this is not 
> relevant:
> I also tried to do something similar and found the ann external quite 
> useful (artificial neural net) - because you can train the net to 
> generate m-dimensional output to n-dimensional input.
> 
> So you train some specific inputs to the desired output values for e.g. 
> a synthesis engine and the other, non-trained input data will be between 
> those trained samples ... this was quite usefull and interesting to play !
yes, neuronal network is a solution. but i never saw anything really convincing (for me), so i did not really use NN.

there is few diferences between NN and the Active Mapping Space (AMS) from previus mail :

the AMS does not nead traning to map n in to m out.
NN could output data even outside of the training scop (and that the main interest of NN for mapping)


anyway the problem with ANN and AMS is how to generated desire input/output value.

cyrille



> LG
> Georg
> 
> _______________________________________________
> PD-list at iem.at mailing list
> UNSUBSCRIBE and account-management -> 
> http://lists.puredata.info/listinfo/pd-list
> 
>