IAIFI Symposium, day two

Today was day two of a meeting on generative AI in physics, hosted by MIT. My favorite talks today were by Song Han (MIT) and Thea Aarestad (ETH), both of whom are working on making ML systems run ultra-fast on extremely limited hardware. Themes were: Work at low precision. Even 4-bit number representations! Radical. And bandwidth is way more expensive than compute: Never move data, latents, or weights to new hardware; work as locally as you can. They both showed amazing performance on terrible, tiny hardware. In addition, Han makes really cute 3d-printed devices! A conversation at the end that didn't quite happen is about how Aarestad's work might benefit from equivariant methods: Her application area is triggers in the CMS device at the LHC; her symmetry group is the Lorentz group (and permutations and etc). The day started with me on a panel in which my co-panelists said absolutely unhhinged things about the future of physics and artificial intelligence. I learned that many people think we are only years away from having independently operating, fully functional aritificial physicists that are more capable than we are.

exciting stars

Stars and Exoplanets Meeting at Flatiron was a delight today. Lachlan Lancaster (Princeton) showed his results on a really interesting object he found in the ESA Gaia data. He was inspired by the idea that star clusters might have central black holes, which might retain a very dense, very luminous nuclear star cluster even after the cluster disrupts. But his search of the Gaia data was so simple: Look for things that are apparently bright but low in parallax (large in distance). Duh! And what he found is a very bright “star” that is variable, shows emission lines, and is above the top of the H–R diagram! The ideas from the room ranged from extremely young star to microquasar to technosignatures (who suggested that?). And the thing is incredibly variable.

But there was lots more! I won't do everything, but I will say that Thankful Cromartie (Virginia) showed data from pulsar monitoring (as part of a pulsar-timing project for gravitational waves). She showed that she can very clearly see the Shapiro time delay in the pulses when they pass by the neutron star that is in orbit around the pulsar. This lets them measure the mass of the neutron star accurately. It is very massive! i think it must be one of the most massive neutron stars known, which, in turn, will put pressure on the equations of state. Beautiful results from beautiful data.

Stitch Fix

I had the great privilege of visiting Stitch Fix today, hosted by Dave Spiegel. I was interested in the company for many reasons, but the main one is that it has a large number of PhD astrophysicists on its data-science team. I learned a huge amount while visiting. Here are some random things:

If you are doing data science to inform or support the decision-making of an employee of your company, it is worth spending a lot of computation on that: After all, the employee is very valuable and expensive! On the other hand, if you are doing data science to directly execute commands (for, say warehousing of goods), you better not get it wrong, because if you have a bug, you could literally move lots of stuff where you don't want it!

If you sell clothing, the lead time between buying clothing wholesale and selling it is long! So you can't quickly or in real time feed back customer preferences into your buying choices. That makes prediction of paramount importance! One thing that really surprised me is that Stitch Fix designs and even manufactures some of its own clothing, so they have unique lines of clothing, adapted to their customers' preferences!

And, obviously (but new to me): Clothing is combinatoric! Even in making a standard button-up shirt, there are myriad few-way decisions about collar, buttons, sleeves, relative dimensions, and so on, such that there is no way in the history of all of humankind that you could make every possible (or even every sensible!) version of a standard dress shirt. That puts a data-science-oriented company like Stitch Fix in a very, very interesting position.

#siRTDM18, day 4

Today was road-traffic day at Real-time Decision Making at Berkeley. Jane MacFarlane (LBNL) and Alexandre Bayen (Berkeley) gave great talks about road dynamics. In MacFarlane's talk I learned that provided (by providers) mobile-phone location information is posterior information not likelihood information. And the priors are outrageously informative (like that every phone is on the midline of a known road!). That is good for the user (the mobile-phone owner), who wants navigation information, but not good for anyone trying to do hierarchical inference over phones or people! This is very related to the issues that Alex Malz (NYU) is working on in cosmology.

Bayen focused on the influence of mobile phones on traffic, which has been immense! As mobile phones have gained traction with drivers, they have driven traffic patterns to a non-optimal Nash equilibrium, where all paths from point A to B take the same amount of time. But these same phones also create crazy new nonlinear dynamics, because all drivers get re-routed simultaneously to a small number of alternate routes when something goes wrong. And it is like a repeating multiplayer game, because each routing company is constantly learning the dynamics induced by all the other companies! But this game is played out in the parameters of a set of differential equations, so it is crazy.

Things would be better if we could find a way to cooperate; this led to great lunch discussions with Josh Bloom (Berkeley). We discussed ways to capitalize on the fact that different drivers have different objectives. No existing apps capture this at all: They all optimize for the triviality of minimum expected travel time!

#GaiaSprint is live; dumb ideas

I spoke at length with Daniel Foreman-Mackey about current projects, and also possible April Fools' projects. It is getting late do do the latter, since (as my loyal reader knows), we take our April Fools contributions very, very seriously. When we do them. One idea is to do some probabilistic modeling of the “Alien Megastructure” Kepler source. We also talked about recent breakthroughs with Bernhard Schölkopf and Dun Wang on doing ultra-crowded-field photometry with independent components analysis (ICA).

At lunch, Andrew Zirm (grennhouse.io) proposed that we start a Dumb Ideas in Data Science meetup. The idea is that so many good ideas are dumb ideas. And so many bad ideas! Anyway, I hope this happens.

In the afternoon, I launched the #GaiaSprint web page and registration information. If you want to hack on the Gaia data the moment it is released, then the #GaiaSprint is for you!

robust fitting, intelligence, and stellar systems

In the morning I talked to Ben Weaver (NYU) about performing robust (as in "robust statistics") fitting of binary-star radial-velocity functions to the radial velocity measurements of the individual exposures from the APOGEE spectroscopy. The goal is to identify radial-velocity outliers and improve APOGEE data analysis, but we might make a few discoveries along the way, a la what's implied by this paper.

At lunch-time I met up with Bruce Knuteson (Kn-X) who is starting a company (see here) that uses a clever but simple economic model to obtain true information from untrusted and anonymous sources. He asked me about possible uses in astrophysics. He also asked me if I know anyone in US intelligence. I don't!

In the afternoon, Tim Morton (Princeton) came up to discuss things related to multiple-star and exoplanet systems. One of the things we discussed is how to parameterize or build pdfs over planetary systems, which can have very different numbers of elements and parameters. One option is to classify systems into classes, and build a model of each (implicitly qualitatively different) class and then model the full distribution as a mixture of classes. Another is to model the "biggest" or "most important" planet first; in this case we build a model of the pdf over the "most important planet" and then deal with the rest of the planets later. Another is to say that every single star has a huge number of planets (like thousands or infinity) and just most of them are unobservable. Then the model is over the an (effectively) infinite-dimensional vector for every system (most elements of which describe planets that are unobservable or will not be observed any time soon).

This infinite-planet descriptor sounds insane, but there are lots of tractable models like this in the world of non-parametrics. And the Solar System certainly suggests that most stars probably do have many thousands of planets (at least). You can guess from this discussion where we are leaning. Everything we figure out about planet systems applies to stellar systems too.