Ned Nufre Sella

21 posts

Ned Nufre Sella

Ned Nufre Sella

@NedSella

The only true one

Katılım Eylül 2016
170 Takip Edilen0 Takipçiler
Lex Fridman
Lex Fridman@lexfridman·
Humanity will win in the long-term. We have always found a way.
English
2.1K
1K
14.1K
929.1K
Yann LeCun
Yann LeCun@ylecun·
@dorian_omi The jury is still out as to which SSL method is best among infomax methods (VCReg, MMCR, MCR2) and distillation methods (I-JEPA, DINO, BYOL).
English
1
0
3
628
Yann LeCun
Yann LeCun@ylecun·
An NYU-CDS blog post about a recent paper by a (very) large cast of characters from Stanford, MIT, NYU, and Meta-FAIR on Self-Supervised Learning (SSL) through information maximization. Unless one uses a preventive mechanism, training joint embedding architectures with SSL leads to a collapse: the representation learned by the system is uninformative, or even constant. A number of methods have been devised to prevent this collapse. One class of methods is sample-contrastive: ensure that different inputs produce different representations. Another class is dimension-contrastive: ensure that different variables of the representation encode different aspects of the input. Both types of methods can be derived from information maximization arguments: make sure the representation encodes as much information as possible about the input. Variance-Covariance Regularization, MMCR, and MCR2 (from Yi Ma's group at Berkeley) are all infomax dimension-contrastive methods. They push the encoder to learn representations of the input that fill up the representation space as much as possible. Paper: arxiv.org/pdf/2406.09366
NYU Center for Data Science@NYUDataScience

CDS researchers @ziv_ravid and @ylecun, et al, advance an entirely new kind of self-supervised learning: Maximum Manifold Capacity Representations (MMCR). MMCR breaks away from traditional SSL methods, though still connects with them via info theory. nyudatascience.medium.com/maximum-manifo…

English
9
40
189
74.9K
Ned Nufre Sella
Ned Nufre Sella@NedSella·
@lexfridman 0MG Lex - you want to be neutral in face of a fascist takeover. In my opinion, a poor judgement from someone I always appreciated
English
0
0
0
5
Lex Fridman
Lex Fridman@lexfridman·
I think that neither Trump nor Harris will destroy America if elected president. Call me crazy, but I think that Trump is not a fascist and Harris is not a communist. I think this is a reasonable rational position, but according to the internet it's insane 🤣 Either way, getting attacked by both sides has been mentally exhausting for me. Perhaps that's the design of the current political climate: anyone with moderate open-mindedness needs to be pushed out in favor of a "battle" between dogmatic extremes. This doesn't seem like the right path toward truth.
English
8.7K
2.5K
39.5K
4.6M
Space Monkey
Space Monkey@SpaceMonkey0009·
@SawyerMerritt The stock decline is all a ploy by the left to attack Elon for his support of Trump. Pathetic attempt.
English
6
0
1
942
Sawyer Merritt
Sawyer Merritt@SawyerMerritt·
RBC Capital's Tom Narayan on Tesla: "Elon doesn't think about things on a day to day or month to month basis; He's looking at things from a years if not decades perspective. Eventually we'll get this catalyst with the next-gen vehicle; Nobody's talking about energy storage; If you guys have ever used FSD. It's the most amazing product I've used since the iPhone."
English
167
510
4.7K
1.2M
30RoE
30RoE@30RoE·
@simongerman600 I agree. Shameful, but not even close to what the Romans, Mongols, Aztecs, Arabs and many others did.
English
3
2
49
5K
Simon Kuestenmacher
Simon Kuestenmacher@simongerman600·
One of the most shameful chapters in human history.
Simon Kuestenmacher tweet media
English
176
517
2.3K
284.6K
Jüdische Allgemeine
Jüdische Allgemeine@JuedischeOnline·
Die Zivilisten in Gaza sind nicht unschuldig: Wenn es so etwas wie kollektive Verantwortung für Verbrechen gibt, dann trifft dies im Falle des 7. Oktober zu. Eine Klarstellung von unserem Gastautor Tobias Huch. juedische-allgemeine.de/meinung/zivili…
Deutsch
2K
268
1.2K
1.3M
François Chollet
François Chollet@fchollet·
Reminder that we could have solved climate change at any point in the past 50 years via obvious, economically profitable choices that would also have improved your life quality (mostly by building nuclear power plants and developing urban cores with great train/subway networks)
English
32
126
1.2K
126.8K
Paul Pantea
Paul Pantea@abelian_cat·
@fchollet Most Romance languages call it "double V" I think. Even among the Germanic ones, English might be the exception.
English
1
0
1
311
François Chollet
François Chollet@fchollet·
In English, the letter W is pronounced as "double U", but in French it's pronounced as "double V". Because it's literally two Vs next to each other. An actual "double U" would look like this: ɯ I guess you could also have named it "upside-down M" but that's way more chaotic
English
26
18
290
59.4K
Ned Nufre Sella
Ned Nufre Sella@NedSella·
@francoisfleuret As a German resident in austria you have to book a reservation in the ambassy 3-4 months ahead, get treated by staff like a criminal, and have to wait another 6-8 weeks for delivery
English
0
0
0
149
François Fleuret
François Fleuret@francoisfleuret·
The benefits of 🇨🇭. You decide to have your passport renewed. In 5min you find how to book a rendezvous at the biometric center, and book one for 48h later. When you get there you wait 5min, and it takes 10min. And you get your passport by mail 3 days later.
English
23
3
106
49.4K
World of Engineering
World of Engineering@engineers_feed·
The temperature on the moon at the Apollo 11 landing site was 200ºF (93ºC).
World of Engineering tweet media
English
341
422
7.2K
1.9M
Filip Piekniewski🌻 🐘:@filippie509@techhub.social
With shutting down perfectly fine nuclear power plants in a middle of an energy crisis and hostility with nearby exporter of energy is yet another proof that Germans are world masters in brainwashing themselves into mind-blowing stupidity. (apologies to my German friends).
English
4
3
36
2.9K
François Fleuret
François Fleuret@francoisfleuret·
Are there some clear principles / foundational theoretical results regarding fine-tuning of models? Best practices? It is an important missing piece in my course.
English
10
3
61
33K
François Chollet
François Chollet@fchollet·
Is there a German name for the kind of automation that creates more work than it saves for the humans operating around it?
English
29
20
352
115.3K
Alfredo Canziani
Alfredo Canziani@alfcnz·
What is a ‘couche’? It's *not* an Italian word and I've never seen it before. Any Italian (or French speaking) who could comment on this? Thanks!
Alfredo Canziani tweet mediaAlfredo Canziani tweet media
Trieste, Friuli-Venezia Giulia 🇮🇹 English
30
2
30
47.3K
Dario Perkins
Dario Perkins@darioperkins·
I remember an equity strategist who told me he had a "fool proof" way to win at the casino. Bet on black every time and don't leave until you win. Often think about this 😆
English
14
3
75
44.7K
MIT CSAIL
MIT CSAIL@MIT_CSAIL·
What's your WORDLE starting word?
English
68
4
50
0
François Chollet
François Chollet@fchollet·
The fact that hand-written code has to be read and understood by humans is a powerful regularization term that pushes it towards simplicity. Code written by a search process (whether biological evolution or program synthesis) has no such constraint.
English
20
50
567
0
James Stephenson
James Stephenson@ICannot_Enough·
Here is my latest $TSLA forecast thread (20 tweets), updated to include actual Q2 deliveries & other forecast changes. 4 tweets summarizing my forecast 11 tweets with updated charts 5 tweets with charts that didn't change, for those who haven't seen them /1
James Stephenson tweet media
English
37
100
650
0
Ned Nufre Sella retweetledi
Eric Topol
Eric Topol@EricTopol·
While the United States has made a U-turn, India is on a very worrisome trajectory
Eric Topol tweet media
English
184
2.2K
3.7K
0