Information field theory is pretty cool. (For spatial/spatiotemporal data) There’s a lot of cool stuff about it, but one interesting feature is that it automatically deals with irregularly spaced measurements. It can basically act as both data imputation & predictive modeling in one go.

!!!!

I wonder, has anyone explored chaos in recursive iteration of gen-AI models? E.g., I’ve just been sent this reddit post as a joke www.reddit.com/r/ChatGPT/s/... But humor aside, that’s pretty obviously a classic case of chaotic behavior of a recursively iterated nonlinear map.

After a bit more study on the topic, I must correct myself re: Solomonoff induction. It’s not quite that an Occam’s razor optimizes prediction, but rather, the cumulation of errors in inductive prediction, when using Solomonoff’s prior (which is an Occam’s razor), is bounded -

Letter from the SSE/SSB/ASN councils on the scientific understanding of sex and gender, pls RT

In case this isn't on folks' radar, now is the time to download your favorite publicly available USG resource/dataset. You have until noon on Monday. #econsky #energysky #climatesky

"Let us do away with the job market and mathjobs.com and reference letters. Let us stop pretending we understand each other’s respective sub-fields. ... Let us seclude ourselves in mountain caves and daub mysterious equations in blood across rock faces to ward off outsiders."

Keynote speaker Ossi Mokryn studies interaction-driven contagion models on temporal networks #NetSciX2025

Studying E. T. Jaynes is insane, because, like… As far as I can tell, the man pretty much solved what is the true fundamental approach to non equilibrium thermodynamics in the 80s, and it seems everyone just... Forgot about it?

This post & its part I predecessor form an interesting duo. Part I relies on fundamentally stochastic models, and Part II on deterministic computation, and has some severe criticism on the use of stochastic methods to describe emergence. 1/

This 2024 I’ve been humbled by conversations on #ComplexityThoughts and its (unexpected) growth. From the risks of new pandemics to the promises of AI, from climate change to evolution, I explored many systems through the lens of #complexity science balancing the technical w/ the accessible. 🧵1/

New blog post: “Untangling the hairball using statistical inference” skewed.de/tiago/posts/...

Starting to learn a bit about functional integrals (e.g. Feynman path integrals used in QFT), but all the sources I can find are either very handwavy or limited to quadratic lagrangians, can anyone point me to a good mathematical introduction to functional integrals? #MathSky #QuantumFieldTheory

It’s real. We’ve been redoing all of our advance directives, estate docs, etc.

"Non-White scientists appear on fewer editorial boards, spend more time under review, and receive fewer citations" www.pnas.org/doi/abs/10.1...

I abandoned this line of enquiry to focus on my current research interests (see bio), but this preprint, while some of it needs rewriting, does have what I believe were some really neat mathematics I've developed. #Math #Topology #ExteriorCalculus doi.org/10.48550/arX...

I’m learning statistical inference, and as I learn I’m coming up with my own ideas, as I do, but in this case I end up finding out that almost all of them thus far have been done by E. T. Jaynes before

Once you get to a certain level of baseline knowledge in a field, learning new stuff in it gets a whole new dimension of fun where you get to dissect all the assumptions & constructions that go into the new thing as you learn it One of the ways academic research can be a bit addictive, honestly

When creating PDFs, avoid using "Print to PDF." A screen reader user may still be able to access the text of PDFs created this way, but heading structure, alternative text, and any other tag structure will be lost. Using "Save As" or "Export" can preserve these tags.

Shannon information theory has been readily & very thoroughly applied to physics & other sciences, because it directly deals in the predictability of stochastic information sources - very useful for optimal prediction, inference, etc.