Sunday, January 10, 2016

Recurrent neural nets

Super cool blog post demonstrating the "unreasonable effectiveness" of recurrent neural networks, which allow you to map 1-or-many input vectors to 1-or-many output vectors. Which lets you generate Shakespeare.

Can't wait to see what Susan Athey et al. come up with for economic applications.

And to Andrej: if you don't name your kids Jerin and Alessia, you've got some 'splaining to do.

No comments: