Inventing Modern Sequence Models as a Music 320 Project

Talk overheads by Julius O. Smith III presenting modern sequence models (leading up to the Transformer) from a digital-signal-processing point of view, as they might be developed in a Music 320 course project.

First presented at the CCRMA Open House on May 17, 2024, with later, longer versions for the Stanford WCML meeting (summer 2024) and the DSP Online Conference (October 2024).

Read the slides (rendered PDFs)

No need to build anything — rendered versions are tracked in pdf/:

• CCRMA Open House 2024 — the original short talk
• WCML 2024-08-08 (as presented) — expanded version
• DSP Online Conference 2024-10-22 (handout) — latest and longest version

Building from source

The slides are LaTeX (dsp2seq.tex and included sections), built with the Makefile in the usual latex/bibtex cycle. Figures live in eps/, fig/, png/, svg/.

License

See LICENSE.