About Me


I am a fourth-year PhD student in Department of Statistics at Columbia University. I am advised by Professor Andrew Gelman. More details of my academic background can be found on my homepage.

About This Blog

Given that this world has already been overwhelmingly brimmed by a second-order-increasingly large number of peer-reviewed articles, writing or even reading a web-based blog is not among the most efficient and promising activities one could possibly conduct. Why should I write and why should You read?

I do not have a rigorous proof for this question. Nevertheless, let me start with a motivating story1:

In the seventh century, a prominent Buddhist monk named Xuanzang travelled all the way from the Far East to the Indian subcontinent, and essentially spent decades there as a graduate student2 on Mahayana Yogachara and another several decades translating and spreading it. When being asked why he had been so dedicated to leaning as well as to sharing what had been learned, Xuanzang wrote the following justification3:

Am I not aware that I cannot gauge the firmament from a telescope len?
Am I not understanding that I cannot measure the ocean with a ditch of calabash?
No, No siree! It is my sincerity and passion, no matter how inconsequential, that I am not willing to abandon.

Xuanzang did not put it explicitly, but the underlying math behind his reasoning is that, the convergence rate of the Monte Carlo integration is $(\sqrt{S})^{-1}$, independent of the dimension4. As a result, even though we live in such a messily booming era whose VC dimension is apparently more than infinity, it is still possible to approximate the whole universe to an arbitrary accuracy with a practically reasonable convergence rate, just by randomly writing or reading one blog post every day.

Agnosticism is real, but we could still actively5 learn? Advertisement of preliminary results are pretentious and irresponsible, but rejection sampling is not completely useless (in particle filtering)? Not all blogs are envisaged to be meaningful, but we would only need no more than 0.234 of them to be readable in the first place?

It reminds me of a poem by Dorothy Parker (with slight modification 6):

Agnosticisms pain you;
Active-learning is slow;
High-dimensions stain you;
And uncertainties cause cramp.
p-values aren’t lawful;
Lassoes give;
Jobmarket smells awful;
You might as well read this blog.

Anyhow, if you have feedbacks for any of my blog posts, please do not hesitate to contact me7.

  1. I am not suggesting there are many lessons to be learned from this one data point as that would be against the Bayesian philosophy on induction. This single example sounds more for model checking. 

  2. I admit that I could not exclude the possibility that he might be a visiting scholar (J-1 visa) rather than an F-1 student. But I suppose H1B was the leastly likely option as any lottery had in principle not been legal before the birth of Kolmogorov. 

  3. I translate this sentence myself, and I should clarify that Buddhist monks may seldom utilize emphatic forms in negative sentences. The Google Translate page of it, however, reads, “Not known inch pipe peep days, small sea discretionary Inference can not [sic] do anything; but not discard this micro Cheng [sic], clothing is taken path” — Yes… but did it work? 

  4. Technically speaking, posts from one blog site would not be independent, the effective sample size is thus smaller than the nominal numbers, in which situation the claim does not hold. 

  5. Conceptually, writing a blog is inefficient from an active-learning perspective, because the blogger is more likely to only query areas that he or she has been familiar with. You could also view this problem as auto-correlation from a sampling perspective or over-fitting from a prediction perspective or a piranha from a zoology perspective. A quick remedy is to have more interactions from the audience (you!). 

  6. It turns out that I occasionally dispatch some more vacuous rhetorics of mountebanks and charlatans to another more personal blog (partially in Chinese) . 

  7. In this situation and if you also happen to be within a topological neighborhood of me, or Morningside height, with a euclidean radius of 10 miles, we could also discuss it over espresso and toast Bayes and Stan.