A very simple snarXiv-like fake ML paper name generator.

This is basically just a single notebook that trains a GRU based on previous paper titles, then generates fake ones based on a prompt.

The outputs were originally used for a pub trivia where people try to distinguish real NeurIPS titles from fake ones... they couldn't.

Code samples and the idea for how to do this were shamelessly nabbed from this blog post and the Tensorflow tutorial on text generation with an RNN so credit goes to them.

Install all requirements in a Python virtual environment using the provided requirements.txt.

For the original work, this was trained on the Kaggle dataset All NeurIPS (NIPS) Papers. Of course you're free to train on anything you like.