I just wrote a python program to convert decision trees into neural networks.
It was really fun and I hope to put up the code onto github once I do some analysis of how effective (or ineffective) it is.
The idea is pretty much described in this paper: https://www.cise.ufl.edu/~arunava/papers/clnl94.pdf
My hope is to apply this to a random forest and then train each resultant net in order to get better performance than random forests.
Tomorrow I'm going to do some analysis of how effective it is, but this isn't quite my top priority right now, since I'm taking the GRE on Tuesday