EECS 183 NLP
Published:
Archive
Notes and Cheat Sheet
Review
I wasn’t originally enrolled in this class, but managed to snipe a spot at the last second and joined a few weeks late. Anyways, I had very little actual ML experience going into the semester and got very spooked by the first few lectures. Nevertheless I resisted the urge to drop, and soon learned about attention + transformers for the first time (good timing, because they would soon show up in all of my other classes as well!). There was also a small unit on speech (Noisy channel model) which I appreciated.
Prof. Suhr was nice and lectured well, but I felt the class spent a disproportionate amount of time on linguistics (her research is also quite linguistic centered). In OH she did mention that the grad version, CS 288, was more transformer heavy so I’ll definitely have to take it next semester.
EECS 183 was a pretty reasonable class, 3/5 in difficulty and 4/5 in fun. I most appreciated its coverage of LLMs, such as word embeddings, RNNs, self-attention, transformers, and the training lifecycle. I got nearly top score on the midterm and did decently on the final.
