There's a problem though. By clipping the tails, this approach introduces significant (statistical) bias. Indeed, as we analyze in a paper with @vkerdos @thejonullman (
https://
arxiv.org/abs/2002.09464), we get minimax rates for estimation by *balancing* the bias & variance from noise. 3/n
@thegautamkamath
-
Balancing Bias and Variance in Statistical Estimation Methods
By
–
-
Private Mean Estimation: Clipping, Noise, and Differential Privacy
By
–
By now, we understand private mean estimation pretty well, especially in 1D. It turns out that one of the simplest algorithms, taking the empirical mean of the clipped samples (to restrict sensitivity) and adding noise (to introduce privacy) works pretty well. 2/n
-
Bias-Variance-Privacy Trilemma in Statistical Estimation
By
–
"A Bias-Variance-Privacy Trilemma for Statistical Estimation," with @argymouz
, Matthew Regehr, @vkerdos, @shortstein
, and @thejonullman
. https://
arxiv.org/abs/2301.13334 Private estimators MUST be biased! 1/n -
Non-uniform endorsement thresholds across arXiv research areas
By
–
I believe endorsement ability is highly non-uniform across areas. You may need to post 5 papers in one area, versus 1 paper in a different area, and perhaps even 0 papers in some areas? No clue about the thresholds since even I have better things to do…
-
Conference Acceptance and Research Self-Assessment in AI
By
–
You and me both. What does being accepted to a conference mean? It means something for your career and 3 random reviewers liked your work. But 1 paper means little for my career now, and I have the confidence to already know when my work is good or not.
-
arXiv Endorsement Requirements for Paper Submission
By
–
If you don't have an academic email, and you don't have sufficiently many arXiv papers in a subject area, then you can't submit to that subject area. You need to have someone endorse you before you submit, which is what this allows you to do. More here: https://
arxiv.org/help/endorseme
nt
… -
Necessity versus tradition in technology practices
By
–
Thank you. You're not wrong. Some discussion of that point here. The short is that some of it is necessary due to scarcity, and some of it is tradition (which maybe we could also do away with). https://
x.com/candrianillbor
n/status/1619073745703804929
… -
Oded Goldreich on computational complexity and theoretical foundations
By
–
Pointing again to another passage from Oded Goldreich (
https://
wisdom.weizmann.ac.il/~oded/on-award
s1.html
…) -
Flawed Award Systems in Academic Research Evaluation
By
–
The above was stated for awards to people, but also holds for awards to papers. But even worse, since these decisions are often rushed under time constraints, and may not be with sufficient expertise (which is why I also don't like "top p% awards" like spotlights/orals). 8/8
-
Awards in Academia: Inefficiency and Opportunity Costs
By
–
There's more harms, like the amount of time wasted on nominating and evaluating that could be spent on more valuable things. Oded Goldreich states his oppositions to awards much more eloquently than I can. Screenshot of my favourite passage. https://
wisdom.weizmann.ac.il/~oded/on-award
s1.html
… 7/n