
Asst Prof @CMU_StatDS interested in infectious disease (@CMUDelphi), genomics, time series, and discrete stable distributions.
Pittsburgh, PA
Joined April 2009
- Tweets9.2K
- Following2.3K
- Followers1.8K
- Likes1.1K
divide by n: biased but consistent maximum likelihood estimator. divide by (n-1): unbiased restricted maximum likelihood estimator. If n is big they are basically equivalent. The difference is more dramatic with covariance estimation where high dimensional features cause problems
Why do we divide by n-1 instead of n when we want to estimate the variance? This is Bessel's correction term, and it made no sense to me back in university when I first learned it.
But I finally understood why!
Got 2 minutes to spare? Here's the explanation!!
5
404
For count data with overdispersion, the negative binomial (NB) distribution is often used. It has variance=mean+r*mean^2. But quadratic variance function does not imply NB. The Poisson-lognormal distribution has the same variance function as NB but heavier tails.
9
686
Will Townes retweeted
I'm looking to recruit PhD students for Fall 2027 interested in probabilistic machine learning methods and AI for science, especially in domains such as chemistry, physics, and biology.
Apply to the Cornell CS PhD program & list my name on your application.
Due Dec 15! Please share. More info below:
Our department @CMU_StatDS is hiring! Tenure-track assistant professor position. Apply here: apply.interfolio.com/196645 It is a great place to work. #statistics #datascience
Will Townes retweeted
There is something really concerning (and disappointing) going on in the US right now.
A dangerous and preventable situation.
Let me explain.
nextstrain.org/avian-flu/h5n…
1/
Will Townes retweeted
My time at the Gates Foundation/IDM recently wrapped up! I'm looking for senior/staff roles in AI for science, probabilistic inference, or scientific software — in Seattle or the SF Bay Area.
Also open to consulting.
Here's what I've been working on:
Our family visited @ALCOSAN (regional wastewater treatment plant) open house last weekend, it was awesome! We stayed all day and still didn't see everything. Touring the plant gave me a different perspective in our AI-anxious era...
2
5
286
I do worry about things like increasingly frequent and sophisticated cyber attacks, but scenarios like fully automated biology wet labs controlled by AI or wastewater treatment plants completely controlled by AI seem much less plausible in the near future.
1
1
84
Will Townes retweeted
TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity
Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission?
medium.com/@TmlrOrg/asking-a…
Will Townes retweeted
If you are working on an important bioinformatics tool, I encourage you to apply to our AI for science program for Claude credits. Link in 🧵
Will Townes retweeted
Do spatial transcriptomics foundation models (SFMs) actually learn biologically meaningful spatial patterns? 🧬
Excited to share our new work assessing if current spatial foundation models learn multi-scale spatial variation in gene expression (spoiler: mostly no)
Will Townes retweeted
One of the major pains of academia is having to click through labyrinths of webpages to submit papers.
Now, thanks to @Josephmrich (and AI), problem solved!
His tool is called PaperPush and is available at github.com/pachterlab/paperp…
It's super easy to use!
1/🧵
Will Townes retweeted
Arguably the most boring step in genomics is the first one: normalization. Settled science. Scale + log. Move on.
Except that here's been a huge blind spot in the field. And it matters for AIxBio. A 🧵about what I think may be one of the most important papers I've written. 1/
Will Townes retweeted
A flaw in Person-Δ may be overstating progress in single-cell perturbation prediction models. Pearson warned about this in the 19th century: reusing the same controls induces spurious correlation. Split the controls, and much of the claimed prediction power fades. Link below 👇
TIL that you can use *negative* penalties in ridge regression, and sometimes that is even the optimal choice! jmlr.org/papers/v21/19-844.h… by @hippopedoid #statistics Bayesian interpretation: Gaussian prior on regression coefficients where the variance is negative, "density" like a U
3
242
Will Townes retweeted
If you use dim. reduction, you may be interested in two recent preprints we've posted on contrastive PCA:
The Rayleigh Quotient and Contrastive Principal Component Analysis I & II
w/ @MariaCarilli & Kayla Jackson. They cover a lot of ground from theory to practice. 1/🧵













