Expositions in Natural Language Understanding

And investigations in related fields.

I organize some of my reading notes on this site, with the hope that it reinforces my own understanding.



A bit more about me here.


The design and markup is replicates distill.

After several renditions, I have landed on this approach. Previously, I used Sphinx, rst, and jupyter notebooks. The two reasons I tried using Sphinx was that I was able to directly render jupyter notebooks and I hoped the rst would be more natural write. But while jupyter notebooks are useful, they aren't the best medium for exposition. Furthermore, the rst ends up increasingly complex when trying to import visualizations, code, etc. Creating the expositions with raw html is less natural -- but favorably trades effort for control and flexibility.

I use Adobe XD CC for diagrams, D3 for visualizations, and PyTorch for implementations.