Twitter | Search | |
Vladimir Iglovikov Jan 8
On my new year flight from Lima to SF, I wrote a blog post on the path from my previous job to the current one. TL;DR => it was an ocean of pain and experience obtained at was very useful 😀 If you like it => 50 claps. If not => 49 😀
Reply Retweet Like
Vaibhav Kumar Jan 8
Replying to @viglovikov @kaggle
"Different teams presented their solutions. I apologized for wasting the time of the audience, showed our two slides. Results: First, second, first. First overall." You're an inspiration!
Reply Retweet Like
Vladimir Iglovikov Jan 8
It looks obvious now that your ML muscles are better developed when you compete with 100-1000+ teams that you have at and not in conference type competitions where you have 3-10. But at a time it was a big surprise to me 😀
Reply Retweet Like
Jeremy Howard
Most academics are still not aware that their insular "state of the art" approaches are far short of what any kaggle grandmaster would do in 24 hours
Reply Retweet Like More
Yann LeCun Jan 9
Most academics spend most of their time coming up with new techniques (and new analyses thereof) for which competitions/benchmarks merely play the role of quantitative metrics. Getting good numbers on a particular practical problem is a different pursuit (a valuable one, too).
Reply Retweet Like
Yann LeCun Jan 9
That's why you'll see interesting papers that have so-so numbers on your favorite benchmark, but say: few labeled samples, no data augmentation, single crop, no ensemble,.... You can always add those tricks to get the numbers up.
Reply Retweet Like
Teemu Roos Jan 10
I find this tweet unnecessarily polarizing.
Reply Retweet Like
Aleksei Tiulpin Jan 10
While it is polarizing, it has quite some truth in it :-)
Reply Retweet Like
Vladimir Iglovikov Jan 8
💯
Reply Retweet Like
Leo Boytsov Jan 8
Aren't these 24 hour results possible largely b/c "academics" paved the path?
Reply Retweet Like
Dmytro Mishkin Jan 9
I am lacking Grandmaster status to be credible, but I agree with . Kagglers are great at solving tasks, but the most often build upon academic things like archs, optimizers, etc. It is the best usage of existing tools - yes. Inventing new ones - rarely
Reply Retweet Like
Dmytro Mishkin Jan 9
Moreover, I'd say that is not about being kaggler or academic. There are hard-working&smart people everywhere, be it at Kaggle, Kaiming He in academia or at fastai. Like in AD&D, 20th level is 20th level, regardless if you are monk, fighter or wizard.
Reply Retweet Like