1/n: We are excited to share that our paper on Chroma, a general purpose diffusion model for proteins, is out today in @Nature!



A couple of my favorite highlights in the 🧵below 👇 nature.com/articles/s4158…
2/n: First up, we have validated Chroma-generated proteins in the wetlab since the preprint.

This data adds to the growing evidence in the literature that diffusion models can create realistic proteins with the desired structure and function.

How often do intervals from popular uncertainty quantification (UQ) methods actually contain the observed value?

@BenKompa, @latentjasper, and I investigated this (known in the stats lit as "coverage") for several popular methods.

Paper: arxiv.org/abs/2010.03039

Details 👇 2/ First, the setup. We constructed 95% posterior predictive intervals for popular methods (SVI, MC dropout, ensembles, etc)

We then measured how often these intervals contained the observed value and the width of these intervals.

A few nuggets from @geoffreyhinton's talk from earlier today at the #ml4h unconference. First up, the distinction between statistics and AI (and presumably ML by implication). Overall, I think these are pretty clean contrasts: On interpretability: It's going to be very hard to explain the contribution of 1000s of statistical regularities that a net uses to produce a decision. Humans are really bad at explaining their decisions (most are post-hoc), so we shouldn't hold neural nets to a double standard