Tony Duan Profile Banner
Tony Duan Profile
Tony Duan

@tonyduan_

7,811
Followers
359
Following
15
Media
259
Statuses

@Tesla | @Stanford

Palo Alto | Toronto
Joined January 2014
Don't wanna be here? Send us removal request.
@tonyduan_
Tony Duan
26 days
Lots of hard work and many late nights went into the making of FSD 12.5, from across the team. Many ideas were simplified and re-worked from first principles. Hope everyone has a chance to try it out. It's a release we're proud of.
Tweet media one
301
503
6K
@tonyduan_
Tony Duan
5 months
Last Fall I wrote a blog post collecting my personal notes to understand diffusion models. To my surprise folks have told me they found it useful, so I'm sharing more widely. (Apologies in advance for any typos or errors).
Tweet media one
9
36
296
@tonyduan_
Tony Duan
5 years
We applied ML methods and found it can improve discrimination and calibration of personalized treatment effects for intensive blood pressure therapy. With Sanjay Basu, @pranavrajpurkar , @DillonLaird , @AndrewYNg .
Tweet media one
0
5
22
@tonyduan_
Tony Duan
4 years
Our work Randomized Smoothing of All Shapes and Sizes is at ICML this week. We show how to best choose randomized smoothing distns, esp for L1 adversarial robustness. Thanks to MSFT collaborators @TheGregYang @edwardjhu @hadisalmanX @ilyaraz2 @jerryzli .
Tweet media one
1
5
22
@tonyduan_
Tony Duan
4 years
Our work NGBoost is at ICML this week. We combine natural gradient with boosting approaches for parametric (in-distn) uncertainty estimates. Thanks to mentors and collaborators @anandavati @DaisyYDing Khanh Thai @SanjaybMDPhD @AndrewYNg @UnibusPluram .
Tweet media one
0
3
19
@tonyduan_
Tony Duan
6 years
Countdown Regression: probabilistic survival forecasting inspired by meteorology, applied to predict time to death from EHR data on 3 million patients! Work with @anandavati , @drnigam , @AndrewYNg .
Tweet media one
0
3
14
@tonyduan_
Tony Duan
9 years
This is a great way of thinking about how we choose to tackle problems. @michael_nielsen
Tweet media one
0
6
11
@tonyduan_
Tony Duan
5 years
Great paper comparing MLE vs CRPS for forecasting. Differences arise due to mis-specified data generating process (too light-tailed or heavy-tailed), and diagnosis can be done via calibration curves (W-shape or M-shape PIT).
Tweet media one
0
3
7
@tonyduan_
Tony Duan
4 years
@dennybritz This is excellent. I'm a big fan of this paragraph in particular.
Tweet media one
0
0
7
@tonyduan_
Tony Duan
5 years
@Cmrn_DP Thanks for this library! I actually use it for almost all my research projects, which involve a lot of time to event analysis.
1
1
5
@tonyduan_
Tony Duan
3 months
0
0
5
@tonyduan_
Tony Duan
4 years
@yaringal @arram wrote a blog post with some examples trying to few-shot learn such uncertainty.
0
0
4
@tonyduan_
Tony Duan
2 years
@Tim_Roughgarden Chow-Liu algorithm, super satisfying.
0
1
3
@tonyduan_
Tony Duan
8 months
@tim_zaman Congrats Tim, it's been a real honor!
0
0
3
@tonyduan_
Tony Duan
7 months
@RemiCadene It was fun working together, best of luck in Paris!!
1
0
3
@tonyduan_
Tony Duan
5 years
@andrey_kurenkov Totally in the same boat — I can never really understand a paper until printing it out!
0
0
3
@tonyduan_
Tony Duan
7 years
@AndrewYNg Thanks for tweeting our work!
0
0
3
@tonyduan_
Tony Duan
6 years
@kapeli If possible, I’d prefer to have dark mode for the rest of the OS but light mode for Dash! Makes docs much easier to read.
1
0
3
@tonyduan_
Tony Duan
9 months
@dpkingma @TimSalimans @poolio @hojonathanho Amazing, thanks for the swift reply!
0
0
1
@tonyduan_
Tony Duan
5 years
@hardmaru Dinh et al's NICE (cheating a bit though since it was a workshop paper)!
0
0
3
@tonyduan_
Tony Duan
3 years
@judyhshen Holy shit COMMUNIST wtf LOL
0
0
3
@tonyduan_
Tony Duan
7 months
@eigenhector The trampoline built-in Clojure function comes to mind.
0
0
3
@tonyduan_
Tony Duan
4 years
@raziehnabi ICML takes place awkwardly right in the middle of these 15 days, too!!
0
0
2
@tonyduan_
Tony Duan
9 months
@dpkingma @TimSalimans @poolio @hojonathanho Looks like some details in Appendix D of the VDM paper were removed in the update from v5 to v6 on arXiv. Was this intentional, and if so what was the reason? (Great paper by the way).
1
0
2
@tonyduan_
Tony Duan
6 years
@hardmaru I need to do this for my Zotero library.
0
0
2
@tonyduan_
Tony Duan
8 years
Great to see @stephen_wolfram speak at UC Berkeley this evening!
Tweet media one
0
1
2
@tonyduan_
Tony Duan
8 months
@DavidDuvenaud I interpret this type of communication as a beta distribution prior.
0
0
2
@tonyduan_
Tony Duan
6 years
@barakgila Rip wholeheartedly empathize with the insufferable boredom of Palo Alto.
0
0
2
@tonyduan_
Tony Duan
8 years
Predictions from @RBReich , for “when robots take over”.
Tweet media one
0
0
2
@tonyduan_
Tony Duan
6 years
@CarlinLiao Odds ratios even worse for intuition.
0
0
2
@tonyduan_
Tony Duan
8 years
@andylouisqin Holy crap this is awesome -- so excited to see where Photon ends up!
1
0
2
@tonyduan_
Tony Duan
4 years
@reece_akana @hackerrank Holy shit W I Z A R D!!
1
0
2
@tonyduan_
Tony Duan
5 years
@daylenyang End of an era! 😅
0
0
2
@tonyduan_
Tony Duan
5 years
@Srceh @tommy_da_cat @icmlconf @MVickySchneider @lawrennd @meeliskull @IRC_SPHERE Got it. I'm primarily interested in the heteroskedastic setting. Seems like the simplest option is to group predictions by [some discretization of ] (μ,σ) then calculate ECE / calibration plots over these groups.
0
0
1
@tonyduan_
Tony Duan
4 years
0
0
1
@tonyduan_
Tony Duan
9 years
@Davis_Wang i'm going home. because i'm tired.
0
0
1
@tonyduan_
Tony Duan
6 years
@wastyk HAHAHA this is so niche holy shit
0
0
1
@tonyduan_
Tony Duan
7 years
@andylouisqin @Headspace Sounds chill I’ll try it out!
0
0
1
@tonyduan_
Tony Duan
7 years
@kevinchen I totally remember living through this! LOL
1
0
1
@tonyduan_
Tony Duan
6 years
@wyndwarrior throwbackkkkkk
0
0
1
@tonyduan_
Tony Duan
7 months
@YunTaTsai1 Haven't tried but definitely want to!
1
0
1
@tonyduan_
Tony Duan
4 years
@vicfromdollops HAHAHAHAHA I’M DEAD
0
0
1
@tonyduan_
Tony Duan
10 months
@benjamin_bolte @DrYangSong Wow, you've implemented the latest from the Improved Consistency Models paper already that was fast!!
1
0
1
@tonyduan_
Tony Duan
7 years
@byrook1e Love your music, keep it up!
0
0
1
@tonyduan_
Tony Duan
8 years
@andylouisqin Wow, such a great read!
1
0
1
@tonyduan_
Tony Duan
7 years
@chocolatapp After searching for years, I've *finally* found the app that's replaced Textmate as my go-to editor. Thank you.
0
0
1
@tonyduan_
Tony Duan
7 months
1
0
1
@tonyduan_
Tony Duan
1 month
@karpathy Congrats Andrej, looking forward to learning!
0
0
1
@tonyduan_
Tony Duan
6 years
@andrey_kurenkov This is a great idea!
1
0
1
@tonyduan_
Tony Duan
5 years
@frejohk @chalmersuniv Congratulations! I've learned a lot from your work and presentations in causal inference.
0
0
1
@tonyduan_
Tony Duan
9 years
@casualeffects Great work on Markdeep — look forward to continuing to use it!
0
0
1
@tonyduan_
Tony Duan
7 years
@levie This is the most relatable content I've ever seen.
0
0
1
@tonyduan_
Tony Duan
6 years
1
0
1
@tonyduan_
Tony Duan
5 years
0
0
1
@tonyduan_
Tony Duan
5 years
@reece_akana Love how it came together, great work! 😍
1
0
1
@tonyduan_
Tony Duan
4 years
@leeren Just wanted to say your OSC-52 vim copy/paste over SSH trick changed my life -- thanks!
1
0
1
@tonyduan_
Tony Duan
8 years
@shyupuff Holy crap congrats Austin! :)
0
0
1
@tonyduan_
Tony Duan
7 years
@bradfordcross I could not agree more with this post.
0
0
1
@tonyduan_
Tony Duan
7 years
@Typora Can't believe I only discovered Typora today -- it's the best note-taking app I've ever used!
0
0
1
@tonyduan_
Tony Duan
8 years
@JackSchaedler "Seeing circles, sines, and signals" is so cool -- awesome work!
1
0
1
@tonyduan_
Tony Duan
6 years
@andrey_kurenkov @gradientpub My take: a @distillpub "paper" is intended to be cited, whereas a "blog post" often accompanies a paper that's meant to be cited instead of the blog post.
1
0
1
@tonyduan_
Tony Duan
1 year
@dpkingma Great paper! Will code be released?
0
0
0
@tonyduan_
Tony Duan
7 years
@NIH Rip 😢
0
0
1
@tonyduan_
Tony Duan
9 years
@BenedictEvans British Library!
0
0
1
@tonyduan_
Tony Duan
8 years
@justindross Super interested! Opendoor has been one of my dream companies since last August.
0
0
1
@tonyduan_
Tony Duan
4 years
0
0
1
@tonyduan_
Tony Duan
9 months
@tim_zaman LOL wtf
0
0
0
@tonyduan_
Tony Duan
9 years
@shyupuff Cool! :D
0
0
1
@tonyduan_
Tony Duan
8 years
@dexicality holy crap you're alive
1
0
1
@tonyduan_
Tony Duan
6 years
@ayyyhuang HAHA glad you enjoyed it!
0
0
1
@tonyduan_
Tony Duan
5 years
@barakgila Holy shit dreams coming true!! 😅
0
0
1
@tonyduan_
Tony Duan
5 years
@barakgila LOL yeah definitely
0
0
1
@tonyduan_
Tony Duan
4 years
@reece_akana Spoken like a programming God
1
0
1
@tonyduan_
Tony Duan
9 years
@t___zheng woah you're on twitter! i waste like all my time on this site
0
0
1
@tonyduan_
Tony Duan
3 years
@barakgila Holy shit, glad you’re okay!!
0
0
1
@tonyduan_
Tony Duan
9 years
@andylouisqin reminded me of this, which is always worth a re-read.
1
0
1
@tonyduan_
Tony Duan
4 years
Rich Hickey's A History of Clojure is very well-written.
Tweet media one
0
0
1
@tonyduan_
Tony Duan
7 years
@danwwang Depends on your definition of “serious”. Numbers-wise, 678/3240 acceptances at NIPS this year would make my estimate around 10000.
0
0
1
@tonyduan_
Tony Duan
4 years
@barakgila @BrianKempGA what in tarnation...🤦‍♂️
0
0
1
@tonyduan_
Tony Duan
4 years
0
0
1
@tonyduan_
Tony Duan
4 years
@OmniGroup Is OmniFocus 2 still available for download on iOS? Broke my phone and had to buy a new one, and now I can't find it on the App Store.
1
0
1