[Gneuralnetwork] An interesting paper - not yet published, but preview is online

[Ray Dillinger]

Here is an interesting thing, currently under review for publication.

https://arxiv.org/pdf/1611.01578v1.pdf

I hope (and recommend) that this paper should be accepted at ICLR.

Designing Neural Nets is hard.  Neural nets can learn to do things
that are hard.  Why not train a network to design networks?

This is particularly interesting as a "baby step" toward
self-bootstrapping AI.

                                        Bear

signature.asc
Description: OpenPGP digital signature