Daily Note: 8 November 2026

Today I completed my Digital Signal Processing assignment and submitted it. I spent too much time polishing up the last 10%, but old habits die hard I guess. In any case, I greatly enjoyed it: I learned about DTMF signaling (now I know where those tones come from!) and applied some deconvolution techniques to recover an astronomical image convolved with a point spread function modelling the aperture of the lens. I wont discuss it more, but it was great fun seeing the maths and computation blend together to reveal that which looks to be lost! There’s an existential metaphor there somewhere…

I read through Understanding Diffusion Models: A Unified Perspective by Calvin Luo (2022) which I greatly enjoyed, and it reminded me of this great series on ODEs and SDEs for machine learning that I read through a while back. If I get the time next term, I’d love to sit in on some lectures on stochastic processes and stochastic differential equations.

I read through some other papers on diffusion models, but didn’t enjoy them as much. One thing I find particularly annoying in the ML world is the callous disregard for notational consistency. It’s bad enough sloppily abusing notation, but at least do it consistently! Here vectors, scalars, random variables, random variates… everything is overloaded.

I didn’t cover as much ground as I’d like today (see above re perfectionism). I’ve still to finish my DSP notes or make a proper start on my other two assignments due. I need to work out how to better “cut my losses” and move on rather than get obstinate about finishing something perfectly.

Until tomorrow.