Henri Schmidt
@henrismitch
cs phd student | algorithms | phylogenetics | computational biology
Cannot emphasize enough how much of an own-goal it is to tell a university with a $53 billion endowment and a a giant PR team that you want to destroy them. Trump's goons blew it. Today's news is full of sympathetic profiles of crucial Harvard health research. Lawyers next. 1/
provided the parameters of an discrete HMM, it is well known that can exactly compute the likelihood of the HMM in O(n) time, where n is the number of nodes of the HMM. is there any relevant literature on sublinear time algorithms for approximately computing this likelihood?
linear regression is to machine learning as maximum matching is to algorithms.
It's clear that many do not understand what @NIH-funded research does to improve health. It's time to revive a study published 10 years ago that provides incredible information about this. link in the comment Every single new drug approved by the FDA from 2010–2016 was built on…
any other ideas? this seems unexplored...
while it is standard to use EM to fit the parameters of a HMM, it is not obvious (at least to me) that this is necessarily the optimal procedure. for example, by differentiating through the likelihood computation, one easily computes the gradient of the likelihood. 1/2
GASTON, our method to learn “topographic maps” of gene expression, is out now @naturemethods! IMO the coolest part is a new model of *spatial gradients in sparse data*. As is typical for bio papers, it’s buried in Methods, but see below for a quick outline on the math 👇
Gene expression topography analysis by GASTON portrays domain organization and spatial gradients of gene expression and cell type composition using spatially resolved transcriptomics data. @uthsavc @benjraphael @PrincetonCS nature.com/articles/s4159…
while this may be true more broadly, von neumann’s work is especially relevant to computer scientists! for example, the duality theorem for linear programming and the birkhoff-neumann theorem have found applications in my research, whereas i know little about einstein’s work!
My hot take is that Grey Tribe weirdos ultimately like Von Neumann because Einstein is famous, and so by saying "I know the secret, smarter guy", you get to perform some kind of esoteric Smart Guy knowledge. It's intellectual hipsterism for compulsive contrarians.
when doing math, i strongly suggest *not* working with a sense of urgency.
urgency is a life hack you need to be unreasonable when it comes to "not being able to do a thing" like, anything at all. any reason someone says "you can't just do this" you need to feel extreme urgency to do it you always need to work with urgency
Just saw a new result on the arxiv: polylog approximation for directed steiner tree! arxiv.org/abs/2412.10744 . Amazing breakthrough in a fundamental approximation algorithms problem. Congrats Bundit!