likely llc

So I’ve decided we’re renaming the “blogosphere” to the “blogograph,” since this is more accurate.

Out of undergrad I met Daniel Kessler, PhD, (https://www.dankessler.me/) yes, at Michigan at Skeeps over a beer. Love this guy, he’s a great technician and I adopted his computational workflow. After talking he invited me to work with the University of Michigan Psychiatry Department. Since I was a Bayesian, and learned a lot of computational statistics, MCMC, and the like, he’s like, we have work for you. My introduction to stat comp and Bayesian was from Prof Yves Atchade, PhD while he was at Michigan. So I was hired to translate Eunjee Lee, PhD, at UNC Chapel Hill in Statistics in Hongtu Zhu’s group, to translate one of her models from MATLAB to C++. C++ is faster. The PI was Dr. Chandra Sripada, he was chill, but no blog.

I completed this work successfully. The clinical goal was to be able to forecast brain network connectivity from neurologal disorders such as Alzheimers, mild cognitive impairment and a healthy control. And there are more applications, you can throw in whatever clinical covariates you can. A bit more on this below. This was never published. This was a paid gig.

The problem was, is that Gibbs samplers don’t scale and this was an extremely high dimensional parameter space, and perhaps not well identified, but I’d have to read more of “Big Mike’s” stuff (https://betanalpha.github.io/), because SVD based methods aren’t compatable with HMC, something about a Stiefel manifold? IDK, I’m just trying to survive. Like there’s no direction to go in when you’re optimizing over a sphere, I don’t know.

My Work at the University of Michigan Psychiatry Department in 2017, Part 1

Turns out, it was too slow to be practical, and I got fired. I’m CC’ing the paper below. Anyway, I got picked up by some hedge fund guys from UBS that are still doing well. Eunjee I’d clean this up for you, if you want to publish it.

But anyway, here’s Eunjee’s site: https://eunjee2.wixsite.com/eunjee (and mine, https://likelyllc.com/). Last time we spoke, she works at KAIST (Korea Advanced Institute of Science and Technology).

I was ready to be her student, but she got knocked up by her husband in North Carolina, got hired as a research assistant professor at Michigan’s biostat dept, and then dipped back to Korea.

She invited me to apply to KAIST, where I could be her PhD student, but again, the problem is recommendation letters and administration. To complete the appliciation, I had to send a handwritten letter to Korea, which required me sending a package to Finland (where I asked Prof. Aki Vehtari, PhD) and then he had to send it to Korea in two weeks. Expensive and nearly impossible.

But yeah, I got fired from https://medschool.umich.edu/departments/psychiatry, and got picked up by a spin off hedge fund start up, who were entrepenuers, developers and scientists and formerly worked at https://www.ubs.com/us/en.html, and other companies. Their hedge fund is still alive, but they don’t even have a website… but anyway…

Implementation here: https://github.com/drezap/blgrm_cpp

Let’s get fired, everyone.