Max Welling
@wellingmax
Followers
32,898
Following
434
Media
62
Statuses
1,209
Explore trending content on Musk Viewer
Northern Lights
• 128057 Tweets
オーロラ
• 128041 Tweets
太陽フレア
• 100854 Tweets
#aurora
• 93463 Tweets
Knicks
• 77032 Tweets
#SmackDown
• 62283 Tweets
Pacers
• 52137 Tweets
#solarstorm
• 41725 Tweets
Game 4
• 41380 Tweets
NAYEON
• 40523 Tweets
Brunson
• 33973 Tweets
Ushuaia
• 31904 Tweets
Melanie
• 26209 Tweets
Gaay Hame Palti Hai
• 22442 Tweets
Protection Campaign
• 21355 Tweets
Nembhard
• 16112 Tweets
Panthers
• 15253 Tweets
Hughes
• 14994 Tweets
Donte
• 14081 Tweets
バチコン
• 10342 Tweets
I am super excited to announce that I will be joining Microsoft Research (MSR) as a distinguished scientist on September 1, 2021.
67
77
2K
After two good years at Microsoft Research AI4Science, I am very excited to announce that as of this month I have, together with Chad Edwards, co-founded a new startup in the field of molecular and materials discovery.
59
75
1K
Even after 22 years it still hurts to get your favorite papers rejected. I recall the first paper on ICA I submitted to NeurIPS that got mercilessly rejected. I felt awful! To all those rejected NeurIPS authors: don’t despair, time will average out the wrinkles! Hang in there!
13
79
1K
Congratulations to dr. Thomas Kipf who just successfully defended his thesis through zoom. He graduated CUM LAUDE!!
15
21
708
Just watched Michael Bronstein's keynote at ICLR. It's done so perfectly well that it's almost scary. Absolutely amazing talk that everyone should see.
15
41
709
I have noticed that the “review process” is very similar to mcmc: the further away from the current state you propose the less likely it is to be accepted”.
12
40
700
I wrote this response to Rich Sutton's "The Bitter Lesson"
18
232
699
Can we stop the “X is all you need” meme in ML. It literally says to stop thinking about everything else, and that’s just counterproductive.
33
44
691
And while
@GaryMarcus
and
@ylecun
are debating who claims what and who references who enough, I watched a brilliant talk by Gabor Csanyi on how deep learning is revolutionizing chemistry.
12
27
683
I am humbled and honored to learn that our paper received the Test of Time Award at ICML. It is wonderful to see how a simple idea can grow into a rich research field. Thanks to Yee Whye for being the best collaborator one can wish, and to the committee for choosing our paper.
🧵We are pleased to announce the ICML 2021 paper awards. Congratulations to all authors!
Test of the Time Award from ICML 2011:
• Max Welling and Yee Whye Teh
📜Bayesian Learning via Stochastic Gradient Langevin Dynamics
2
130
828
32
26
669
A simple framework that unifies many well known MCMC algorithms. If you read this you can clean up your brain to make place for new stuff (useful for when you are over 50).
Our
#icml2020
paper "Involutive MCMC: a Unifying Framework" is now available on arxiv . It describes many MCMC algorithms from a single perspective.
Work with
@wellingmax
,
@eeevgen
, Dmitry Vetrov
4
103
437
1
137
635
I can really get used to giving talks over Zoom. No travel, no lost time, no air pollution and you can record and share your own talk! Here is the talk I gave "at" MIT CSAIL's Embodied Intelligence group.
4
88
629
I am extremely excited and proud to share with you the ELLIS PhD program. To students around the world: please apply. This will hopefully become one of the most competitive PhD programs in the world.
11
223
610
Proud to share the webpage for the brand-new Microsoft Research Lab Amsterdam where we will work on molecular simulation for green technology and healthcare. We are hiring deep learning for computational chemistry researchers and engineers.
9
76
527
Please share. This man will be remembered as a hero who didn’t abandon his people.
.
@ZelenskyyUa
's tv address to the Russian (!) people might be the most moving speech that I've ever seen in my entire life. The whole world needs to see, understand and share this crucial Ukrainian message.
#StandWithUkraine
#Ukraine
#Україна
#Russia
#Россия
4K
130K
329K
4
83
510
Today, we officially announced the AI4Science initiative at MSR, where we combine ML with scientific computation to develop faster and more accurate algorithms to simulate physical, chemical and biological processes, with impactful applications to health and sustainability.
Microsoft Research announces AI4Science, a new global team of machine learning, quantum physics, computational chemistry, molecular biology, fluid dynamics, and software engineering experts working to tackle important societal challenges. Learn more:
6
174
722
3
44
508
I gave a talk on Wednesday in the Feynman Lecture Hall where Feynman taught his famous Feynman lectures. It was a humbling experience.
8
18
471
Congratulations to superstar Dr. Taco Cohen for graduating Cum Laude yesterday.
13
10
467
I learnt this today: “All steel produced after 1945 is contaminated with background radiation because of the use of nuclear weapons. Such steel is unusable for many scientific and medical applications. Steel made before 1945, eg taken from sunken battleships, is be used instead.”
8
34
457
This is an amazing 271 page paper by Maurice Weiler on Gauge Convolutional Neural Networks on Riemann manifolds. Absolutely masterpiece by this truly talented AMLAB PhD student. And beautifully illustrated as well. I am so blessed with these wonderful students around me!
Happy to announce our work on Coordinate Independent Convolutional Networks.
It develops a theory of CNNs on Riemannian manifolds and clarifies the interplay of the kernels' local gauge equivariance and the networks' global isometry equivariance.
[1/N]
24
325
1K
2
60
438
I am super excited to announce this really exciting work by an amazing dream team on molecule generation using equivariant flows. It's the first equivariant model that simultaneously generates atom locations and discrete properties and beats the state of the art by a good margin
Very excited to share our latest work E(n) Equivariant Normalizing Flows for Molecule Generation in 3D. Joint work with
@emiel_hoogeboom
@FabianFuchsML
@IngmarPosner
@wellingmax
.
Paper:
5
63
327
2
50
408
I also want to speak out against reducing the accepting rate. A rejected paper is very discouraging for student (Note: I am not saying we should accept bad papers). If necessary create a new category of only papers only presented by short video.
9
40
403
MSR will open a new office in the beautiful city of Amsterdam where I will help assemble a diverse team of researchers and engineers to work on the problem of molecular simulation. I am thrilled to announce we have already made our first hire: Rianne van den Berg!
12
27
394
Our paper on Functional Neural Processes is now online:
(it's a NPB deep learning architecture without GPs)
2
66
347
It is with gratitude that I announce my resignation from Qualcomm. I joined Qualcomm in August 2017 after the acquisition of Scyfer and have spent the next four years with fantastic colleagues to build up Qualcomm AI Research. 1/n
12
2
352
16 experts were interviewed on global warning. All said that it will be above 2C. 60% thought between 3 and 5. At 4 degrees large parts of the planet are inhabitable by the end of the century. Your children and grandchildren will experience the resulting chaos. Think about that.
15
46
339
And here it is: the long awaited bible of geometric deep learning. Great job!
Proud to share our 150-page "proto-book" with
@mmbronstein
@joanbruna
@TacoCohen
on geometric DL! Through the lens of symmetries and invariances, we attempt to distill "all you need to build the architectures that are all you need".
More info below! 🧵
26
534
2K
2
48
327
If you were using a graph-CNN on a meshed surface, stop right there. You are ignoring the geometry. Use mesh-CNNs instead! They look like graph-CNNs but are more powerful, because the kernel is not necessarily isotropic.
Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs
by Pim de Haan et al. including
@maaurricce
,
@wellingmax
#ComputerScience
#Learning
1
16
56
2
63
299
People have the tendency to dig their opnion trenches and start shooting at the perceived enemy. This happens a lot in religion, but also in science: multiverse vs collapse, bayesian vs frequentist, and scaling versus reasoning. Get out of the trenches people!
12
17
288
We are starting a group in neural PDE solving and are recruiting the project lead. Interested to lead a group of 5-10 researchers and engineers on this topic? Please apply now!
4
65
279
All papers and video's for the ELLIS workshop on geometric deep learning are now available on the website.
0
57
265
Merging VAEs with Flows into SurVAE flows. Great work by Didrik Nielsen who was on a highly successful ELLIS exchange from Denmark TU at U. Amsterdam. (w/ E. Hoogeboom, P. Jaini, O. Winther).
2
46
263
I watched two more tutorials today: one on structured priors and one on quantum ml. Both amazing. Highly recommended.
6
17
250
And we are worrying about rogue AI while the world burns. The is a very predictable existential risk for humanity that will cause femine, large migration, and subsequent societal instability and war.
BREAKING: Vietnam 🇻🇳 and Laos 🇱🇦 have both set a new national heat record today. This comes just 3 weeks after Thailand 🇹🇭 measured over 45°C for the first time in its history, setting a new national heat record.
Absolutely brutal.
311
5K
8K
13
46
242
Please join us this Friday April at the 24 at the ELLIS Geometric Deep Learning virtual workshop. We have an amazing lineup of speakers and the invited talks will all be streamed. Thanks to Thomas Kipf and Erik Bekkers for the organization.
1
54
239
This is what true leadership looks like. My thoughts go to the people of Ukraine, to the young soldiers on both sides who are asked to give their lives for a senseless war, their parents who live in fear, and everyone in Ukraine and Russia who stand up to this madness.
‘I have asked 27 European leaders whether Ukraine will be in NATO .... Everyone is afraid, no one answers,’ Ukrainian President Volodymyr Zelenskiy said in a late night speech
1K
8K
22K
0
20
241
Robbert Dijkgraaf, theoretical physicist at the University of Amsterdam and director of the Institute for Advanced Studies at Princeton will become the new minister of science and education in the Netherlands for D66. I wish him all the best in politics!
1
21
231
My talk at the Institute for Advanced Studies (IAS) on July 21, 2020.
1
51
218
Sindy strikes again with an oral at Neurips. The truly amazing thing for me was that if you train a VAE unsupervised with addional internal phase degrees of freedom it will automatically align phases within objects and segement them. 🫨
Excited to share Rotating Features – accepted as oral at
#NeurIPS2023
Rotating Features learn to represent object affiliation via their orientation on real-world data without labels.
Let’s dive in!
📜
🖥️
2
87
508
2
19
218
I completely agree. It’s more effective to support an excellence network and encourage mobility. I am very worried this CERN of AI will costs loads of money that can be much better spent and in the end does not attract the kind of top researchers you want to have there.
"CERN of AI" is a terrible idea.
we don't need publicly funded AI moonshots, it's impossible to compete with
@DeepMind
/
@OpenAI
/
@facebookai
in that game.
we need to develop and retain talent more effectively. the
@ELLISforEurope
plan is the right answer.
11
23
146
6
27
211
It will only cost 2% extra investment of our annual GDP to stay below 1.5 degrees climate change. That’s around 1.7 trillion a year. We spend 1/3 of that on direct subsidies to fossil fuels. Tax Carbon emissions and that to invest in sustainability!
4
28
213
Nice piece in Quanta about gauge CNNs and geometric deep learning
1
67
208
My talk at the ECCV Workshop on "Perception Through Structured Generative Models.
1
45
207
Nobody seems to be talking in my bubble about that war that’s coming within a week or so in Ukraine. Is it just me that’s worried?
29
4
204
My latest blog on Bayesian bits, qubits, and..... Hintons (the neural bosons in a quantum neural network!).
Check out the latest blog from VP of Technology
@wellingmax
—he examines the question of how to adaptively determine the required precision for an AI prediction task, exploring work from
@Qualcomm
#AI
Research that uses probabilistic bits and quantum bits:
0
13
67
4
28
196
Congratulations to Christos Louizos who successfully defended his thesis today and has joined the ranks of “doctors”.
4
3
194
Last Tuesday
@maurice_weiler
graduated from U. Amsterdam Cum Laude with his 500 page thesis "Equivariant and Coordinate Independent Convolutional Networks". Huge congratulations! Thanks to the stellar committee (also incl.
@KyleCranmer
,
@gfbertone
,
@lipmanya
,
@DaniloJRezende
)
21
11
195
It’s time to add a 6th G to the 5 G's of geometric DL: Grids, Groups, Graphs, Geodesics, Gauges and Geometric Algebra. (
@TacoCohen
,
@mmbronstein
@joanbruna
,
@PetarV_93
)
Congratulations to a fantastic team:
@jo_brandstetter
,
@rejuvyesh
,
@vdbergrianne
.
1
15
189
Until now, I was unsure what to think of Elon Musk. After all, he is responsible for accelerating the introduction of electric cars and commercial space flight. But this tweet does it for me. He is nuts and dangerous. I stand with Fauci.
4
3
190
What a soothing oasis of rest to find one of Gabriel Peyré posts on math in between all the sceaming and mud throwing on X. Thank you.
Fourier transform on groups turns convolution into multiplication. For non-commutative groups, it is matrix-matrix multiplication though …
6
150
983
3
11
182
If you apply to this MSR funded position at EPFL you get to work with Bruno Correia
@befcorreia
, Michael Bronstein
@mmbronstein
and myself
@wellingmax
on equivariance for modeling proteins. Please apply!
4
34
180
@yudapearl
It’s unlikely that data-driven ML will be replaced. Rather it will be augmented by causal modeling. And I bet that even causal modeling will eventually become data-driven itself in the form of causal generative models.
2
19
181
MSR opens new office in Berlin! Amazing to now have
@FrankNoeBerlin
as a colleague! Super excited to start working together on projects in sustainability and health to make a positive societal impact. If you also feel inspired, make sure you apply to our job openings!
As we advance Microsoft Research's AI4Science initiative, we’re opening a lab in Berlin. Learn more from Chris Bishop, Technical Fellow & Director, & Frank Noé, our new Partner Research Manager, as they discuss the fifth paradigm of scientific discovery.
5
63
318
1
10
178
I am hiking and what do you know: there is a Deepmind trail :-/
6
1
173
Bayesian Bits: a new method to learn (I mean infer from your approximate Bayesian posterior) the number of the bits required in your neural network. When no bits survive you prune.
1
45
172
In this paper formulate transition path sampling as a path integral optimal control problem, which closely relates to a Schrodinger bridge problem. Fantastic work by a great team!
Excited to share our latest work at the intersection of machine learning and computational chemistry; Path Integral Stochastic Optimal Control for Sampling Transition Paths between molecular conformations. With
@YuanqiD
*,
@priyankjaini
, Ferry Hooft,
@BerndEnsing
and
@wellingmax
3
59
319
0
20
168
I think this is important. I understand the EU stops its funding of Russian research grants, but let's be careful not to transfer blame to our Russian colleagues. Many are also victims, risking their lives to bring change. Maybe that's our best hope.
3
14
163
The next generation of fully steerable E(3) equivariant GNN architecture. The paper also presents a neat unified view on equivariant message passing methods. Brilliant work by
@jo_brandstetter
,
@robdhess
,
@ElisevanderPol
,
@erikjbekkers
. (Johannes is on ELLIS exchange from Linz)
Are you interested in graph structured data?
Do you want to include geometry and physics to boost your GNNs?
Check out our paper on Steerable E(3) Equivariant Graph Neural Networks.
Joint work with
@robdhess
@ElisevanderPol
@erikjbekkers
@wellingmax
5
97
454
2
32
159
Despite the nay sayers always pointing out problems I can not help but be impressed. And yes, there must be some level of understanding to complete this task. (But no worries, I am not a member of the scaling club).
3
6
158
In my 2018 keynote at ICML I showed this curve and predicted that in 2025 we would have models with 100 trillion parameters. We might get there sooner...
7
28
157
Tomorrow, Tu March 2 2021, we will have an ELLIS Workshop on Deep Learning for Medical Imaging. Please sign up at the link below if you want to participate or simply watch the live streaming (link on webpage). See you tomorrow!
3
27
154
I am co-organizing MLSS-Indo 2020 in Indonesia. . Indonesia is a vast country with around 6,924,511 students but little papers in ML. I am looking for sponsors who want to support a student to attend the summerschool. Contact me if you are interested!
6
50
152
Congratulations Chris. It was a magnificent thesis and defense. I recommend your thesis to everyone who wants to learn about topological deep learning!
🚀Today, I successfully defended my PhD thesis on Topological Deep Learning
@Cambridge_CL
📖.
I was honoured to discuss my work with my examiners,
@wellingmax
and
@jmhernandez233
. Very grateful to everyone who has been part of this journey and in particular to
@pl219_Cambridge
!
77
148
2K
1
6
150
ELLis is holding a workshop on how the ML community can get involved in CORVID19 research on April 1st. Open to all.
1
63
144
My talk at GIS. GIS was a wonderful conference in Paris that I can recommend to anyone. I learnt a lot about information geometry and optimal transport!
1
18
145
Interesting statement by Roald Hoffmann. While this is what happened to CV and NLP, it may not happen like that to chemistry. ML will become a very powerful tool for the comp. chemist, but data is expensive and the microscopic eqns. are known! So inductive bias will remain key!
A provocative statement by Roald Hoffmann at the closing of
#WATOC2020
#quantumchemistrydemise
17
81
435
9
13
140
My blogpost on a possible path to AGI.
Hybrid architectures are key to designing
#AI
models of the future. Learn about neural augmentation in this insightful post from our very own Professor
@wellingmax
.
0
14
47
3
15
139
Back from 3 weeks of teaching ELBOs and Diffusion Models at AIMS in Cape Town. What an inspiring class of students! Thanks to
@ulrichpaquet
and Claire David for hosting me!
0
6
142
That was the biggest crowd I have ever given een talk to....(ICML Stockholm)
2
13
137
June is International Pride Month. For the occasion Andy Keller trained a Topographic VAE on Rainbow MNIST. I warmheartedly support the LGBTQ+ community!
1
3
137
Can somebody please implement an autopilot recommendation letter generator:
F: [CV,score]-->Recommendation Letter. I am struggling writing recommendation letters on Sunday afternoon while the sun is shining outside and there is a huge pile of other burning issues catching fire..
12
5
136
The human brain is far to good for the purpose it has evolved. Intelligence might therefore suddenly emerge through some kind of phase transition at some level of complexity. Might something similar have happened to GPT-4? I find it too intelligent for how it is trained.
27
6
130
A gif is worth a thousand papers
Isn't it strange that, despite decades of work in generative models, denoising diffusion was developed so recently?
9
24
318
0
17
135
Brilliant work by Andy Keller on unsupervised learning of approximately equivariant capsules.
Together with
@wellingmax
, we think deep learning needs more organization and structure... topographic organization and equivariant structure 😁
Introducing our new paper:
Topographic VAEs learn Equivariant Capsules
📃
🧬
1/6
3
86
475
2
15
134
The value in academic research is in its amortization: a good idea can find its way into millions of systems even though for any particular one it is only a small part of the puzzle.
Interesting thread. Solving real world problems usually involves a huge amount more than just applying an AI method. In my experience the AI / ML part is often not the hardest part of a deployed AI based system.
5
44
311
2
25
131
@yudapearl
This book is for Russian researchers. Most of them are against the war. Why starve them from intellectual food? We should not alienate them I feel. In the Cold War scientists kept collaborating.
6
4
132
Sindy Löwe acing her oral presentation. Congratulations to her and Peter O’Connor and Bas Veeling for the honorable mention. (Note that Sindy was a master student when she wrote this paper!)
0
7
131
Like to work on cutting edge generative AI using physics tools such as the Renormalization Group, with the briliant mathematician & physicist prof Miranda Cheng at U. Amsterdam (co-supervised by me)? Think no more and apply to below!
2
23
125
More exciting work on geometric DL. This time we bring Geometric Clifford Algebra;s to deep learning.
New work on Geometric Clifford Algebra Networks (GCANs). We propose geometric templates for modeling dynamical systems. A 🧵on geometric / Clifford algebras, and symmetry group transformations in neural networks.
📜
7
100
428
0
17
128
My brilliant student Emiel Hoogeboom starts his talk with the epic words: "Work with the brilliant Rianne and also quite clever Max Welling". No further comment.
2
6
127
R-paper = 3: For each paper I help submitting I have around 3 ideas for new papers. But there is only resources available for 1. That’s why the real R value is very close to 1.
4
5
125
This is the recording of a public talk I gave in the Copenhagen Koncerthuset. It was an amazing experience to talk to a sold out concert hall before an audience of >1000 or so. Thanks to Jacome Armas for making this possible.
0
16
122
Another stunning blogpost by
@maurice_weiler
to explain his new book (which you should read if you want the ultimate intro to equivariant DL).
This is the 2nd post in our series on equivariant neural nets. It explains conventional CNNs from a representation theoretic viewpoint and clarifies the mutual relationship between equivariance and spatial weight sharing.
👇TL;DR🧵
1
65
310
0
24
126
I hope I didn’t overreach with the last sentence of my abstract, but I will make the best of it 😝
7
6
121
We are hiring a full prof in theoretical computer science. Among the suggested topics are foundations of AI and quantum ML. Please apply!
0
48
115
Amazing work by Victor and Emiel on E(n) equivariant graph NN applied, among other things to the prediction on molecular properties.
We are glad to present "E(n) Equivariant Graph Neural Networks". A new simple & effective method to build E(n) equivariance into your graphs. We will present it at ICML. Code is now also available at: . Joint work with
@emiel_hoogeboom
and
@wellingmax
.
7
68
343
0
17
116
While
@GaryMarcus
and
@ylecun
are debating who claims what and who references who (or not), I now also watched a talk by the great Kieron Burke on ML powered DFT. Kieron teaches at
@UCIrvine
where I also taught for 10 years, and we were almost neighbors!
I think this tweet should become a trend. While
@GaryMarcus
and
@ylecun
are debating who claims what and who … I started learning conformal predictions 😀
2
1
40
5
4
119
Nice work by Leon Lang and Maurice Weiler on how to design group equivariant CNNs based on the Wickert-Eckart theorem which is also used in quantum physics. To be presented at ICLR 2021.
How to parameterize group equivariant CNNs?
Our generalization of the famous Wigner-Eckart theorem from quantum mechanics to G-steerable (equivariant) convolution kernels answers this question in a quite general setting.
Joint work with
@Lang__Leon
[1/n]
5
28
129
0
18
118
Amazing talk by Katie Bouman on the process of reconstructing the now famous image of a black hole. There seems be lots of synergy with MRI image reconstruction.
1
11
118
Incredibly cool work by
@emiel_hoogeboom
,
@vgsatorras
and
@ClementVignac
on using equivariant diffusion models to generate stable and pretty decent sized molecules. Much better than previous EGNN flow model. And of course, mind blowingly cool GIF for you to enjoy 🤯.
Molecule Generation in 3D with Equivariant Diffusion (). Very happy to share this project (the last of my PhD woohoo 🥳) and a super nice collab with
@vgsatorras
@ClementVignac
(equal contrib shared among three of us) and of course
@wellingmax
7
68
429
1
21
117
Two Tenure Track Positions in Machine Learning - Universiteit van Amsterdam
1
37
116
What an honor, mentioned in the same sentence as our CEO
@satyanadella
by our CTO
@kevin_scott
. Better than 10 NeurIPS papers. Looking forward to the event!
This Wednesday I'll join
@SatyaNadella
and
@WellingMax
to talk about the future of research at the Microsoft Research Summit. Join us online and celebrate the last 30 years of exciting work from
@MSFTResearch
:
1
34
101
0
5
116
This seems like a dream job to me if you are interested in AI4Chemistry and you are willing to move to Berkeley. It's connected to the BIDMaP Institute that works in AI4Science more broadly for climate change. Very inspiring effort
0
20
113
For the podcast featuring my interview with Chris Bishop, see:
2
11
111
Excellent news!!
The
#MIDL2020
team is thrilled to announce that, thanks to its partners and sponsors, this year's virtual conference will be FREE!
Although registration will be required, everyone will be able to participate in our program for free.
Registration link:
1
196
380
0
7
111
Super exciting news for Thomas Kipf. Congratulations! Well deserved.
Very honored to receive the ELLIS PhD Award for my thesis on Deep Learning with Graph-Structured Representations -- alongside with with
@NagraniArsha
for her work on multimodal DL (congrats!)
38
34
507
1
5
110
We also announced the new location for the Amsterdam team at the top floor of Matrix 1 at Science Park.
If you are inspired, apply to one of the open vacancies!
2
6
112