Tweeps - news! I'll join
@MSFTResearch
as research manager starting Oct 1st, opening a new AI4Science Lab in beautiful Berlin! We'll focus on fundamental challenges between
#MachineLearning
&
#Physics
/
#Chemistry
. We'll start at Alexanderplatz, and we'll hire at all levels! 🧵
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.
I made my introduction to
#DeepLearning
lectures (lots of references to shallow
#MachineLearning
and
#Physics
) public.
Hopefully useful for newbies who want to get an overview of methods before specializing on state-of-the-art stuff. Feel free to use!
Wohoo - PauliNet, our deep
#NeuralNetwork
solution of the electronic Schrödinger equation is out in
@NatureChemistry
. Huge respect for
@jhrmnn
and Zeno Schätzle for championing this. Deep Quantum Monte Carlo is the future!
German future prize for Sahin, Türeci, Kariko and Huber for developing the first approved Covid-19 vaccine and bringing the m-RNA technology to scale with
@BioNTech_Group
. I couldn't think of anything more appropriate. Huge congrats and thank you!
#zukunftspreis
@kkariko
The number of monthly new ML +AI papers at arXiv seems to grow exponentially, with a doubling rate of 23months.
Probably will lead to problems for publishing in these fields, at some point.
Microsoft Research welcomes Visiting Researcher, Prof Frank Noé (
@FrankNoeBerlin
), a leading expert in machine learning for the natural sciences.
Learn more about opportunities across our global labs.
It's finally here: the DeepTime software library!!!
Massive work led by Moritz Hoffmann, with Martin Scherer,
@tmhmpl
,
@andreasmardt
, Brian de Silva,
@brookehus
, Stefan Klus, Hao Wu, Nathan Kutz and
@eigensteve
.
Let's celebrate this seminal paper by Katalin Karikó
@kkariko
. Describes how mRNA can be modified so that it doesn't cause an immune reaction. This is the basis for mRNA to be used in the Covid-19 vaccines of
@BioNTech_Group
,
@moderna_tx
and
@CureVacRNA
.
Stochastic Normalizing Flows: We can train normalizing flows by combining invertible networks and stochastic dynamics (MCMC, Langevin) in any order. Better expressivity and sampling than either one.
With
@jonkhler
& Hao Wu
#MachineLearning
#DeepLearning
Who wants to come to Berlin and work with
@jhrmnn
and myself on the next generation of deep learning systems for Quantum Chemistry, especially deep QMC, exploiting ideas from Stat Mech, generative deep learning, equivariant graph nets etc? Send Email or DM.
Just read this. Beautiful and elegant method for computing chemical potentials by
@ChengBingqing
. This is the first time that a grand-canonical "simulation method" makes sense to me - by simply avoiding to do the simulation grand-canonical at all.
Can you believe it, I have been staring at protein structures for nearly 20 years but just ran the first gel in my life. The luxury of being a computer guy.
2020 won't be less exciting. My personal/private highlights:
-
@CecClementi
will move to Berlin as an Einstein professor for physics
- We are buying an apartment in Schöneberg
- We are getting married!!
Antisymmetry is what makes electronic structure calculations hard. We need more
#MachineLearning
work to make progress with this fundamental problem. Stefan Klus takes a stab at it by developing antisymmetric kernels:
A bright moment in dire times.
@CecClementi
being appointed Einstein professor for physics in Berlin by FU president Günter Ziegler (with safety distance).
@Einstein_Berlin
,
@FU_Berlin
.
First official release of the DeepQMC code for Quantum Monte Carlo with
#NeuralNetworks
wave functions, including architectures like PauliNet, FermiNet and DeepErwin. By Zeno Schätzle,
@PBerntSzab1
, Matej Mezera and
@jhrmnn
Hi Tweeps: Doing a Markov Modeling / Time Series Analysis Zoom tutorial using DeepTime and PyEMMA on Feb 21-24. We will schedule everything from noon to 5 pm CET so that at least US east-costers have a chance as well!
20 spots. Please register - see:
2006: semi-prof reggae band
2008: politics & economics
2009: pure math
2011: major depression
2012: art school
2013: 20h/w webdev
2016: bsc applied cs
2017: married
2018: msc AI
2018: start PhD in ML+Physics
2020: beat depression
2021: DeepMind internship
2023: first real job
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:
Hao Wu has developed a very nice approach to compute reaction coordinates of molecules with normalizing flows by leveraging the idea that in reaction coordinate space the system's time propagation should follow a simple dynamical law.
I have a proposal. How about removing any occurrences of
#MachineLearning
and
#ArtificialIntelligence
from paper and workshop names as it’s trivially an ingredient in pretty much everything now?
Tweeps - apply here for senior researcher and engineering positions in
@MSFTResearch
AI4Science Berlin! We'll keep these open for a few more weeks.
If you have already applied please give us some time to respond.
Tweeps - news! I'll join
@MSFTResearch
as research manager starting Oct 1st, opening a new AI4Science Lab in beautiful Berlin! We'll focus on fundamental challenges between
#MachineLearning
&
#Physics
/
#Chemistry
. We'll start at Alexanderplatz, and we'll hire at all levels! 🧵
🚨ANNOUNCEMENT🚨
The
@covid_moonshot
is now published in
@ScienceMagazine
! The paper reflects the major contributions of the FAH community and a global network of scientists collaborating to make drug discovery open, globally accessible, and affordable
We are setting up an
#AI
/
#MachineLearning
-driven protein design platform. Looking for a highly motivated PostDoc with experience in protein purification and/or RNA biology. Ideal: experience with experimental protein design. Collab with Daumke (MDC) and Schacherl (Charite).
Woohoo, Stochastic Normalizing Flows coming up as a
#NeurIPS2020
spotlight talk. Thanks 10^6 to Hao Wu and
@jonkhler
.
P.S. Arxiv version is identical in methods and results. That version was rejected from ICML. Except a few improvements for camera ready!
Stochastic Normalizing Flows: We can train normalizing flows by combining invertible networks and stochastic dynamics (MCMC, Langevin) in any order. Better expressivity and sampling than either one.
With
@jonkhler
& Hao Wu
#MachineLearning
#DeepLearning
Excited: first joint paper with Klaus-Robert Müller, Alex Tkatchenko and
@CecClementi
, coming up as an Annual Review in Physical Chemistry.
First strike in our efforts to unite Machine Learning for Quantum and Statistical Mechanics
Roadmap on
#MachineLearning
in Electronic Structure: Perspectives on current + future challenges on predicting material properties, learning force-fields, solution of the many-body problem etc.
Pls Share: Wanna do a
#MachineLearning
PhD in Europe? Any core ML discipline or ML + Phys / Chem / Bio?
Apply to
@ELLISforEurope
here by Dec 1st. Work with two top Labs/PIs of your choice.
My group has openings, please contact for more info.
Super interesting work by Falkner, Coretti, Romano, Geissler (RIP Phil) and Dellago: They show how Boltzmann Generators can be extended to help sampling transition path ensembles efficiently.
Working towards computing kinetics for large biomolecules with independent Markov decomposition.
@tmhmpl
championed this work, big shoutout to the team Mauricio del Razo,
@CTLeeRes
@RommieAmaro
@biobryn
.
So happy to have a paper with you Rommie!
There are still honest people in the 🌎.
Lost my backpack with credit cards, IDs, IPad in the chaos of moving with a shared car. Finder researched me and gave it back. Super Happy.
Tweeps, wish you all an awesome, healthy and happy new year! Hope you and your dear ones stay safe, you get rewarded for your hard work, your grants and startups get funded, your AIs get a little smarter.
And don't overdo it with resolutions. Eating is a good thing! Buon Anno!!
Interested in participating in an online tutorial on Markov modeling and using the deeptime + PyEMMA software packages on March 1-3? There are still spots available - please register by Email (see link).
The protein TMPRSS2 is a great COVID-19 drug target, but we have no experimental structure. Here we provide the as-yet most advanced equilibrium structure model and study how two drug candidates, camostat and nafamostat work.
This one was very tough, and it's finally out! In-depth analysis of ubiquitin-SH3 protein-protein binding mechanism combining extensive MD, Markov modeling and NMR with
@smnlssn
, Kalyan Chakrabarti, Christian Griesinger, Thomas Weikl and many colleagues
As someone loving
#MachineLearning
and Physics/Chemistry, I am overwhelmed by
#NeurIPS2020
having 5 exciting workshops on this interface. We are not writing papers that fast...
[1/N] Generative AI has revolutionized how we create text and images. How about designing novel materials? We at
@MSFTResearch
#AI4Science
are thrilled to announce MatterGen: our generative model that enables broad property-guided materials design.
👇
Wonderful work led by
@MohsenSadeghi
on the coarse-grained simulation of biomembranes with incorporating hydrodynamics, realistic kinetics and long timesteps. This is meant to be used in particle-based reaction-diffusion simulations of cells:
Interested in a PhD on the interface of
#machinelearning
and biology/biophysics/comp chem? This is a good opportunity and allows us to admit both students with Bachelor and Master degrees on separate tracks:
@leonklein26
introduces flow matching with equivariance. This allows us for the first time to get flows working for iid sampling of molecular structures in Cartesian coordinates in such a way that we can get reasonable Monte Carlo acceptance rates.
@adad8m
On the risk of getting yelled at, but I would do this in <5 min with PowerPoint. Takes some practice but is super efficient for these things when you know how.
Immense pleasure to having worked with this team and finally see this out. Using
#MachineLearning
, the first CG forcefield that predicts free energy landscapes similar to all-atom for unseen proteins, and it's much faster than all-atom. Tour-de-force by
@CecClementi
+team
Very happy and grateful that Chemistry at
@RiceUniversity
has renewed my adjunct appointment. Looking forward to come back and visit when the Covid situation allows us.
Ho ho ho, now I a have a machine ... actually it's a camera :-) Thanks to
@CecClementi
I have a wonderful new scuba toy - can't wait to try it out 🐠🐟🐡🐬🦈📸
Welcome new lab mate
@SherryLixueC
joining today as a researcher in
@MSFTResearch
AI4Science. Looking forward to working with you on the frontier of AI for quantum Chemistry.
Most downloaded articles in Annual Reviews 2020 includes our review on ML for molecular simulation with
@CecClementi
, Aleks Tkatchenko and Klaus Müller. It's free for download temporarily here (also on arXiv):
@AnnualReviews
@bifoldberlin
Marvelous day with the
@CecClementi
and Noe groups and we managed to sway the weather into admitting a BBQ. A few people were missing but now we have a nearly complete group shot that is photoshoppable to perfection.
We propose temperature-steerable flows that learn a family of probability distributions parametrized by temperature. Allows to use flows in combination with parallel tempering simulation for sampling. Led by Manuel Dibak and Leon Klein.
Opening at
@FU_Berlin
for a three-year postdoc position for
#MachineLearning
in the molecular sciences. Candidates with strong theory and method development knowledge in Quantum Chemistry or Statistical Mechanics are particularly encouraged to apply.
We are building a sklearn-style
#MachineLearning
framework for dynamical models from time series data: MSMs, (E)DMD, VAMP, Sindy, kernel&deep versions of those. Led by M. Hoffmann w.
@eigensteve
,
@brookehus
,N. Kutz,S. Klus & everyone invited!! What's your favorite package name?
We live in a time of crises and uncertainty, but also a time where science has shown to have profound and direct impact on the course of events. I want to dedicate the 2nd half of my career to be part of this, developing key technologies and scaling them to real-world relevance.
New version of excited PauliNet -- a deep Quantum Monte Carlo approach to compute ground and excited states of the electronic Schrödinger equation is online. Now including conical intersections!
Moritz Hoffmann figured out a way to generate valid Euclidean distance matrices without going through the 3D representation. We believe this may be a very useful tool when generating molecules or other particle data.