Recursion and Memory

[richie]

Hi Mihai,

from time to time I get back to MEPX, and always find it fascinating and very useful to easily find relations in data.

I was thinking that, in a sense, the result function given by MEPX is a bit like a trained feed-forward neural network, and in this sense it doesn't have "memory" of past values. Would it be possible in principle - and would it even make sense - to give MEPX a "sense of memory", where it has the possibility to feed-back onto itself the last output, and/or having internal memory in the form of variables which are modified and passed on to the next iteration? So that, in a way, the result would be similar to a recurrent neural network? Of course the main application would be in timeseries prediction, where long and short term relations would be useful. I'm thinking on how good are echo state networks in predicting chaotic systems and was musing if such modified version of MEPX could be used to make forecast in the same kind of systems...

As always, thanks for the good work!!

Richie

[Mihai Oltean]

Hi,

I'm currently adding generalized operators (like sum, min, max etc over multiple values (more than 2)). Then, we will add some recursions and functions.

best regards,

Mihai

--
You received this message because you are subscribed to the Google Groups "Multi Expression Programming" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mepx+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mepx/72659621-29d4-40b3-a924-da97e5c06befn%40googlegroups.com.

[richie]

Sounds great! Looking forward to it!

Cheers,

R.