Grigory Pereleman was making less than $100/month while working on the solution of the Poincaré conjecture at the Steklov Institute. He won't live forever, but his ideas will. newyorker.com/magazine/2006/…
His critiques of mathematicians hold true for scientists more generally: "...there are many mathematicians who are more or less honest. But almost all of them are conformists. They are more or less honest, but they tolerate those who are not honest."
The @NewYorker piece is filled with good quotes and anecdotes that many could learn from.
"Speed means nothing. Math doesn’t depend on speed. It is about deep." - Yuri Burago. This is so so true, and not just for math.
When asked for a C.V. for a job application, Perelman pushed back: "If they know my work, they don’t need my C.V.,” he said. “If they need my C.V., they don’t know my work."
Good point.
On the matter of ethics in science and mathematics:
"It is not people who break ethical standards who are regarded as aliens, it is people like me who are isolated."
While the 2006 @NewYorker piece is insightful in the quotes it presents, it is also revealing in the authors' words. Apparently the opposite of the math profession's "nerdy stereotype" is a person who has "a succession of girlfriends". What the ?#@! is that?
Towards the end of the article you learn that Yau is so green with envy that he apportions credit using the arithmetic 100=50+25+30. Sort of funny, but also sad.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Is a single-cell RNA-seq atlas really an atlas? A short thread about #scRNAseq, maps, and atlantes (yes, the plural of atlas is atlantes! h/t @NeuroLuebbert). 🧵1/
Atlantes must be accurate to be useful, and the vexing question for centuries, namely how to best represent the spherical earth in 2D, is nontrivial. There have been many proposals with pros & cons for each (because the sphere and the plane have different Gaussian curvatures). 2/
In #scRNAseq, atlases of cells have become synonyms with UMAP figures of gene expression matrices (used to be t-SNE but UMAP seems more popular now). Map making from gene expression matrices is more challenging than map making of our 3D world; #scRNAseq is in ~10⁴ dimensions. 3/
The 17 #BICCN@nature papers on the primary motor cortex in mouse (+some human & marmoset) that were published yesterday are a major step forward in terms of open science for an @NIH consortium. For reference, links to the open access papers are here: nature.com/collections/ci… 1/🧵
First, the #BICCN required preprints of all the papers to be posted on @biorxivpreprint, and as a result the papers were already online 1-1.5 years ago. Of course the final versions now published have been revised in response to peer review. 2/
Speaking of peer review, almost all the papers were published along with the reviews. In combination with the preprints, this provides an unprecedented view of how consortium work is reviewed and how authors respond. Real data for this perennial debate:
In 2008, as a new professor of molecular and cell biology @UCBerkeley I presented at a seminar series intended to introduce 1st year students to research in the department. Two profs. presented each time, with food beforehand. I was paired with Thai food and Peter Duesberg. 2/
I knew of Peter Duesberg and his HIV/AIDS denialism, but I hadn't realized that he worked @UCBerkeley. We were now colleagues in the same department. 😱 3/
but I propose an additional platinum standard for one click reproducibility.1/
By "one click", I mean that the entire analysis be reproducible in a (free) interactive online session of @colab (or other similar service). All steps of the analysis, from downloading data to generating figures are then not only automated but accessible for users. 2/
In response to questions & comments by @hippopedoid, @adamgayoso, @akshaykagrawal et al. on "The Specious Art of Single-Cell Genomics", Tara Chari & I have posted an update with some new results. Tl;dr: definitely time to stop making t-SNE & UMAP plots.🧵biorxiv.org/content/10.110…
In a previous thread I talked about the (von Neumann) elephant in the dimension reduction room: t-SNE & UMAP don't preserve local or global structure, they distort distances, and they are arbitrary. Almost everybody knows this but they are used anyway...
There were some interesting technical questions about our work. One question was the extent to which PCA pre-conditioning affects results. We examined this (Supp. Fig. 3). Tl;dr: it's time to stop making t-SNE & UMAP plots (with or without PCA pre-conditioning).
It's time to stop making t-SNE & UMAP plots. In a new preprint w/ Tara Chari we show that while they display some correlation with the underlying high-dimension data, they don't preserve local or global structure & are misleading. They're also arbitrary.🧵biorxiv.org/content/10.110…
On t-SNE & UMAP preserving structure: 1) we show massive distortion by examining what happens to equidistant cells and cell types. 2) neighbors aren't preserved. 3) Biologically meaningful metrics are distorted. E.g., see below:
These distortions are inevitable. Cells or cell types that are equidistant in high dimension must exhibit increasing distortion as they increase in number. Actually, UMAP and t-SNE distortions are even worse (much worse!) than the lower bounds from theory.