Week notes — 1st August 2024

I’ve been away, but I’ve also done a bit of work.

Specifically, the Stanford Natural Language Processing with Deep Learning course. I’m up to assignment 4, which is suddenly much much harder than previous assignments. The assignment asks you to build up some of the key components of an RNN, train it for a couple of hours and get some results.

Where previously I had a hand waving understanding of an RNN, having worked through the assignment I feel like I’ve really started to get a grip of what’s going on. Even though the concept is relatively simple, with an encoder, decoder and attention there are a lot of moving parts.

Having flowed through the course relatively easily, it was frustrating to spend two weeks working through the assignment, however it really did crystallise things for me. Going back through a couple of previous lectures I felt I was understanding things at a much deeper level.

There’s only one further assignment left, so I suspect I’m almost over the hardest part. I’m not sure what I’ll do to simulate the project work that is part of the course.

Completing the course will get me to about 10% of the way through a masters, and I need to start thinking about how I might structure ad hoc learning into a proper structure. I’m constantly tempted by fundamental courses I find interesting (graph learning, fundamentals of probability) , while I suspect the most useful courses to me are more applied.