In terms of impact,
@AndrewYNg
is probably is the most influential AI researcher.
Mostly not because of his top 100 peers. More likely due to his ML courses and the millions of people who learnt from them, and Coursera, which democratized the knowledge barrier to working in ML.
I'm happy to be on the Time AI 100 list of influential people in AI, and thrilled that 8 others from my Stanford group or other teams I led are also named!
Congratulations to my former advisees and team members:
- Sam Altman
@sama
(as undergrad, interned with me on RL
Hinton's Doctoral students: Ilya Sutskever, Alex Krizhevsky, Ruslan Salakhutdinov, Radford Neal, Richard Zemel, Brendan Frey...
Post doctoral students:
Yann LeCun, Peter Dayan, Max Welling, Zoubin Ghahramani, Alex Graves...
This is magic!
So, it turns out you don't get much time to throw together a Nobel Prize party, but we got people together to celebrate! 🎉
Here's me & Geoff before the party started, & a nice group photo of all of Geoff's former PhD students that were able to attend on short notice.
Transformers can now play chess!
"We train a 270M parameter transformer model with supervised learning on a dataset of 10 million chess games. We annotate each board in the dataset with ... roughly 15 billion data points. Our largest model reaches a Lichess blitz Elo of 2895".
Google Deepmind presents Grandmaster-Level Chess Without Search
paper page:
largest model reaches a Lichess blitz Elo of 2895 against humans, and successfully solves a series of challenging chess puzzles, without any domain-specific tweaks or explicit
@ash_natarajan
just because we won without Ashwin, doesn't mean we couldn't have won faster with Ashwin on a turning 5th day track.
Clearly Kohli has some personal issues with Ashwin and he's letting it cloud his judgement.
In CS, these are the top 3 for me:
1. Pagerank co-authored by Rajeev Motwani.
2. Primes are in P by Manindra Aggarwal, Neeraj Kayal, Nitin Saxena
3. LPs are in P by Narendra Karmarkar
Terrific points made and conversation in the quoted tweets.
I would like to add that there are other institutions, other than IISc+older IITs, that I think are fantastic for CS grad studies:
1/ TIFR
2/ CMI and IMSc
3/ IIT Gn and Ropar
4/ IIIT Hyderabad and Bangalore
5/ BITS
To all the prospective PhDs in/from India, before you sign up for a PhD position in the US, please carefully consider if you would do better (PhDing) in India. I sketched my experience in this blog . Feel free to send me your comments 😃
@engineers_feed
simple phenomenon:
A glass vase cracks when it falls to the ground, but one hasn't as yet seen a cracked glass falling back together as a vase.
One purported explanation is that this explains arrow of time, ie increasing entropy indicates forward motion of time. Why?
Wonderful 5 days at COLT2023.
Amongst all the problems (and their solutions) i learnt of, one of the most interesting problems that stuck with me was the unlearning problem. (1/3)
The first three years of my PhD, i mostly worked alone. The pandemic didn't help. I felt research life was mostly an arduous but lonely journey. The sense of researchers being a community was quite missing.
But in the last few months all that has changed.
(1/3)
Recent advisory of
@GoI_MeitY
needs to be understood
➡️Advisory is aimed at the Significant platforms and permission seeking from Meity is only for large plarforms and will not apply to startups.
➡️Advisory is aimed at untested AI platforms from deploying on Indian Internet
@QuantaMagazine
For those who have not read the Arthashaastra, it is a tour de force in real-politik and economics. India's equivalent of Sun Tzu's work.
Didn't know it was lost to history until today. Now waiting for people to (re)discover Gangesa, who wrote tatva chintamani.
Wanna Learn Mathematics for Free?
Here are some of the Best freely available playlists from Linear Algebra to Number theory that you can use to excel at Mathematics.
A Thread 👇
@kaushikcbasu
On brevity - the best paper in biology of the entire 20th century, by Watson and Crick, on the discovery of the DNA structure, was exactly 2 pages long.
You can see it here:
Truly happy that Apoorva Khare has won the Shanti Swarup Bhatnagar award.
My own encounter with his teaching was through Linear Algebra course in which he regularly sprinkled nuggets from category theory.
Amazing researcher, teacher extraordinaire. And now SSB awardee!
Kolmogorov randomly dropping an article about functions in a math magazine for high school pupils in 1970 is peak math culture/education one could dream of.
Important Correction.
I'm told by much better informed sources than me that:
a. CDS department at IISc has ~50 GPUs.
b. Most IIT profs have at least 1 V100.
---
Having said that, the context is that one pre-training run of an 8B model LLM is 1 month at a 500 GPU cluster.
@NTanjore
@PChidambaram_IN
You saying Kaushik Basu's economics is sub-standard is like Donald Trump saying Einstein's physics is sub-standard.
No actually, it's worse.
@MajorPoonia
Why is an army man afraid of facing the truth?
Truth is that everyone in India knows atleast one person who has died of Covid. If each family knows ~ 100 other families, & there are 20 crore families in India about 20 lakh people have died
Truth is truth sir. Can't hide from it
@__Abhimanyu_S
@Memeghnad
yes.
For one, it'll force many congressites to not sit on laurels of having won twitter battles over the last few months and actually take to the streets.
Working with this group was a game changer for me
Each meeting with Karthikeyan Shanmugam, Arun Suggala, Dheeraj Nagaraj and the inimitable Prateek Shenoy lead to some deep insights
Prateek is not just technically strong, but deeply cares about outcomes for members of the group
Excited to share that the Machine Learning and Optimization team at
@GoogleDeepMind
India is hiring Research Scientists and Research Engineers! If you're passionate about cutting-edge AI research and building efficient, elastic, and safe LLMs, we'd love to hear from you. Check
@LuxMeaMundiAM
@Gannuuprem
Actually elephants in india are very well kept. they form close bonds with the humans caring for them. they're definitely better off than in the wild.
Think of this like how you would think of a dog or a cat playing ball.
Yesterday I truly felt part of the small but close-knit CS theory community, as we had researchers from UW, NYU and NUS visiting us at IISc. Grateful for all the fruitful discussions I had with them.
Here's all of us saying hi!
(3/3)
@Rainmaker1973
It's the choice of colours. It's similar to colors you see in color blindness tests, the eyes can distinguish, but only after some staring.
The peripheral vision has fewer cones that distinguish colours, and more rods - which makes the contrast in the image harder to distinguish.
@sgsiyer1970
@srivatsayb
@RahulGandhi
No. It simply means Public pays more
If Adani Ambani pay for lease, where is the money going to come from? Airport User fee goes up. Highway tolls go up. Economy dampens. You start having monopoly for the lease holder
The day will come when they will "Monetize" our water supply
Interested in any of the topics:
CS/AI/ML?
Science Communication? Education?
Comics? Puzzles? Paradoxes? Juggling & Games?
Register for the "Art of Computing" workshop, a unique Walmart CSR Initiative, at IISc on Feb 17-18.
Great food! Excellent talks! Please register (for
"A generation later, all the Sanskrit mathematical texts stored ... were brilliantly summarised by the Persian polymath Khwarizmi, whose name is the origin of our word “algorithm” and whose book ... Kitab al-Jabr is the basis of our word “algebra”.
"
Congratulations to
@DalrympleWill
whose magnificent The Golden Road: How Ancient India Transformed The World is published today.
The depth & range of knowledge is extraordinary; the story told is revelatory.
An immense achievement by a truly brilliant historian.
@DeepikaBhardwaj
@milaapdotorg
With the proposed 1 crore 20 lakhs raised for just four kids to study at Oxford, 100s can benefit through opening of a new college here.
Convex Optimization is the epitome of what
@JSEllenberg
calls "Applied Pure Math." In what other engineering class would you encounter the Hahn Banach Theorem, semialgebraic geometry, and the support vector machine in the same lecture?
@theipaper
"Regular jogging is not associated with increased mortality in men, as shown by the significantly lower mortality in joggers than non-joggers in our study."
Well cited study with 10 times the sample size
Thrilled to share that our work on Causal Attribute control was accepted at
#ACL2023NLP
.
We showcase our work on the use-case of toxicity mitigation and show that our method can be used plug-and-play on any pre-trained language model of your choice.
@kaushikcbasu
@kaushikcbasu
The Govt PR machinery will selectively hear your tweet as
"Kaushik Basu: India's asset monetization sounds good - a plan to unlock govt assets"
Relating to Russian asset capture/oligarchies, a great book is Red Notice:
In this work, we study the causal bandit problem where we identify a near-optimal intervention from a set of possibly non-atomic interventions in a causal graph, utilizing the idea of covering interventions.
Many thanks to coauthors
@ayushsawarni
@BarmanSiddharth
@sinhagaur88
.
Paper titled “Learning Good Interventions in Causal Graphs via Covering” co-authored by Ayush Sawarni, Rahul Madhavan, Gaurav Sinha, and Siddharth Barman has been accepted for publication at UAI 2023
#uai2023
@BarmanSiddharth
On rigour and stages of learning:
"The point of rigour is not to destroy all intuition; instead, it should be used to destroy bad intuition while clarifying and elevating good intuition."
To me scientists change the world because their effect on outcomes is interventional. If they weren't doing science, the world would be different.
Our world would be very different without the four folks in this picture.
Nice talk by
@ketd47
at
@IIScCSA
on "The Sampling Problem Through The Lens of Optimization" where he shows that the sampling problem SVGD is equivalent to the optimization problem of SGLD.
Many professors on this list are from the Chinese universities. I bet many of them are working with recent graduates/undergraduate students for such a large number of publications.
It is rare to see undergrads in India with any research publications.
How to fix this?
@ddale8
@zoo_bear
@zoo_bear
does this day in an day out. In India, all he gets is trolled all day for doing this thankless work.
Good to see him get some international recognition, even if just a share and a thank you!
@blissbyosho
"start a campaign how PM Modi gave so much"
Yea fix what's wrong by doing more of the mistakes made.
Campaign, election management, buying MLA this is what BJP is good at.
Desh tho chalana nai aata.
This was my question at the IISc interview --
Prove that row rank = column rank for any matrix,
Under the constraint that if you get it wrong, you won't get into IISc 😅
@aryehazan
I'll bet most people are surprised when they discover that row space dimension is the same as column space dimension. The proof is simple of course, and once you know what is going on it is not a mystery, but by then you are an initiate and will soon learn the secret handshake.
@pmddomingos
Or don't pick a topic that is hot.
Pick an area, read a lot and then pick a topic that interests you -- one you'll be happy to spend a few years researching.
Computer science has fundamental laws that are universal. If you prove something (and your proof is right), the statement you make is true universally.
So students who like the physics way of science (but don't like where physics is heading) should consider Theory CS.
Using Physics as your template for how science works makes you stupid about science. Physics is a weird little science with a very simplistic causal structure. It’s fundamental laws are universal. They operate always and everywhere. No other science is like that.
Claude3.5 is insanely good for mathematics. No not arithmetic. Mathematics.
Discuss your favorite theorem ideas with it. Ask what supporting lemmas should be. Discuss possible extensions. Ask it to formalize your ideas. Check everything.
spent a few hours with Claude 3.5 sonnet doing some mathematics research. you are underestimating the impact AI will have on research. yes, you. yes, I'm serious. no, it does not replace mathematicians. but the augmentation is about to take off.
I am a sum total of what my teachers have made me and inspired me to do!
From high school to now, their role in shaping my thoughts, my path, cannot be overstated. 🙏
One of the best journalistic pieces you will read all year. Reads like a crime novel.
Well done
@mknid
, wonder how you got so many people to speak up!
@mondalsudipto
@P39A_nlud
Finally passed my PhD qualifiers!
Didn't know one of the answers, but figured it out with direction/hints from the supportive committee.
Is this is true of research life as well? A good researcher doesn't have to know everything, but given effort, can figure it out?
21k run yesterday!
I used to have headaches after my run, but yesterday-felt fresh.
What made a difference:
1) decreasing to a pace I was comfortable with (10.5kph).
2) taking water breaks (once per 5k).
3) hydrating right.For me this is 700ml per 5k. Not much more or less.
@rose_k01
what sentence did he use that sounds like a derogatory remark to you?
Have you just received some mail from IT Cell to target him this morning?
And yet Professors do this on a daily basis - teach students who do not know what they are about to learn.
Communicating what you know is a serious part of understanding the concept. You cannot teach until you have probed the concept in a 100 different ways yourself.
i've met many people much smarter than me and it's very entertaining watching them attempt to communicate with people who don't understand how smart they are
@BarmanSiddharth
wins ACM India ECR Award!
Siddharth is one of my favorite collaborators & a great friend. Apart from being a brilliant theoretician, his taste for problems, attention to detail & presentation, and impeccable academic integrity -- fortunate to learn a lot from him
@BhavikaKapoor5
People are fighting everyday. They're getting their skulls busted, trolled all day, losing friends, yet standing up. In a much smaller way many more people are speaking up in their circles and fighting.
Not all people have lost courage.
@arindamkhan
@neeldhara
can you get GPT-4 to correct the paper as well? 🙂
it should perform poorly, but would be good to see if it gets anything right - i.e. grades any correct answer as correct and a wrong answer as wrong.
Join us in exploring the triumphs & challenges of building a 2.5 Mn sqft marvel with a unique forest belt and architectural wonders.
Watch "Superstructures - The Making of Terminal 2 Kempegowda International Airport Bengaluru" on Dec 21 @ 7PM, on National Geographic.
@BLRAirport
Exactly! Firstly, there are very few degrees in data science. Secondly, some of the best finance people and data scientists i've worked with have had a physics background.
Some of the most effective data scientists I've ever worked with, haven't had degrees -- or had degrees in physics, statistics, biology, history, or linguistics. 🙅♀️
This perspective is so myopic. A person and their capabilities aren't dictated by any academic pedigree.
# on shortification of "learning"
There are a lot of videos on YouTube/TikTok etc. that give the appearance of education, but if you look closely they are really just entertainment. This is very convenient for everyone involved : the people watching enjoy thinking they are
@littmath
Corrected statement for those interested -
"an irrational _algebraic_ power of an algebraic number other than zero or one is transcendental."
A paper titled "Mitigating Bias in Language Model Evaluators: A Causal ATE Approach", authored by Rahul Madhavan and Kahini Wadhawan has been accepted at SIGNLL Conference on Computational Natural Language Learning (CoNLL) 2024.
Hello Everyone,
We have few urgent competitive exams books pending in our Amazon Wishlist, for long now, please consider ordering them.
Here is the Amazon Wishlist Link-
Retweet for Visibility
#SupportCommunityLibrary
#FreeLibraryMovement
#BooksForAll
Algorithms that you discover in your sleep!
I've had two algorithms/analyses that I've woken up and written down.
(albeit quite simple ones as compared with the below cited one).
@CmonMattTHINK
P(F) = 2/3
P(F|T) =4/5
P(F|H) =4/7
---
Info gain on 1 throw~0.2285
P(F|HH)=8/17
P(F|HT)=8/11
Info gain on 2nd throw after H ~0.2566
--
P(F|TT)=8/9
P(F|TH)=8/11
Info gain on 2nd throw after T ~0.1616
--
P(F|HHH)=0.94117
P(F|HHT)=0.8421
Info gain on 3rd throw after H ~0.0990
Now, I speak as much as i can about problems I am working on with my research friends. Get feedback from them about solution method as well as problem setup.
Equally I get to understand their problem areas and what techniques they use in their problems. Ideas get exchanged
(2/3)
if you trust
@GoogleDeepMind
Gemini about itself, it has 1.56 trillion parameters and cost Google $1-2billion (as opposed to GPT-4 which cost OpenAI $500M.) there were more than 100 engineers in the team who worked on Gemini.
jailbreak by 한국어
🤣🤣🤣
Looking forward to
@ketd47
's talk on a related topic next week at
@iiscbangalore
.
Hopefully will get to learn more about this area.
"The Sampling Problem Through The Lens of Optimization : Recent Advances and Insights".
Given a dataset (say of customer data) and a hypothesis (model) class, we can learn a best model instance to fit the dataset.
Now say a few customers leave. You want to unlearn their data points from the model. What can you do?
(2/3)
Writing an entire project code in 3 hours is now reality.
But you still need to:
(1) architect the project.
(2) find the algorithms that make your code fast (LLMs often don't care about time/sample/space complexity).
(3) fix errors that the LLM (not so rarely) makes.
@Geeta_Mohan
@LNajafizada
Blinken: We will wait for actions
Journo: Women are getting beaten. Journalists are arrested. People are raided. Local media is shutdown. Murals Destroyed
Blinken: We will wait for actions. That is very important. We have an enduring commitment to Afghan people
---
Joker US Admin