Jonathan Vandermause

Research scientist @DEShawResearch, previously @Harvard, @dartmouth

[Jonathan Vandermause @jonpvandermause]

Our work on machine learned reactive force fields is now out in @NatureComms!

We show they can be trained automatically and mapped onto polynomial models that are 2x faster than ReaxFF 🔥

Paper: https://t.co/Ml1ypck8TK
Code: https://t.co/tZnzB6ri6c

[Jonathan Vandermause @jonpvandermause]

@zany_cloud Awesome work Claudio! Congratulations! 👏👏👏

[Jonathan Vandermause @jonpvandermause]

@littmath Solution: Let h_{m, N} be the number of outcomes in an N-toss game with m the diff between Bob and Alice's scores, final toss heads. Similar def for t_{m, N}. Easy to show the following recursion relation:
h_{m, N+1} = h_{m+1, N} + t_{m, N}
t_{m, N+1} = h_{m-1, N} + t_{m, N}.

[Jonathan Vandermause @jonpvandermause]

@littmath Here are the probabilities out to N=1000 tosses. At N=100, Bob has a ~48.6% chance of winning, Alice has a ~45.8% chance, and there is a ~5.7% chance of a tie. But Bob's edge diminishes as the number of tosses increases.

[Materials Intelligence Research @ Harvard @Materials_Intel]

FLARE is powering simulation AI on Microsoft Azure Quantum, with AiiDA... Hope we get new materials faster! Thanks to amazing developers at MIR: Jon Vandermause, Yu Xie, Anders Johansson, Steven Torrisi, Lixin Sun, Andrea Cepellotti, Simon Batzner, Cameron Owen, David Lim.

[Jonathan Vandermause @jonpvandermause]

Shoutout to my collaborators @YuuuXie, David Lim, @cameron_cowen and @bkoz37 and to the @Materials_Intel group for making this work possible

[Jonathan Vandermause @jonpvandermause]

Our work on machine learned reactive force fields is now out in @NatureComms!

We show they can be trained automatically and mapped onto polynomial models that are 2x faster than ReaxFF 🔥

Paper: https://t.co/Ml1ypck8TK
Code: https://t.co/tZnzB6ri6c

[Materials Intelligence Research @ Harvard @Materials_Intel]

Reactive ML molecular dynamics of 0.5 Trillion atoms on 27,336 GPUs with autonomous active learning + uncertainty + record speed + first principles accuracy that matches catalytic activity measurements https://t.co/9lBHTSKv02 @OLCFGOV

[JCTC Papers @jctc_papers]

[ASAP] Multitask Machine Learning of Collective Variables for Enhanced Sampling of Rare Events https://t.co/PukK6gU1nS

[Claudio Zeni @zany_cloud]

We apply FLARE mapped force fields (@Materials_Intel
) to fit fast-and-accurate machine learning potentials, and we adopt unsupervised machine learning to characterize the "phase" of atoms at the surface and in the inner part of Au nanoparticles at different temperatures. 2/4

[Claudio Zeni @zany_cloud]

It's a pleasure to share with you our latest article on Au nanoparticle melting appearing in @NatureComms https://t.co/fGJjwpAR0Q…
A fantastic collaboration with @_RossiKevin, T. Pavloudis, J. Kioseoglou, S. de Gironcoli, R. E. Palmer, & F. Baletto. 1/4

[Volker Deringer @vl_deringer]

Published now in @ACSChemRev - "Gaussian Process Regression for Materials and Molecules", an introduction to GPR #MachineLearning methods as applied to #compchem & materials science. Openly available at https://t.co/ppKNRwCO6F (1/5)

[COSMO Lab 🇨🇭🇪🇺 @lab_COSMO]

After a tremendous amount of work - by @vl_deringer and @apbartok in particular - it is a real pleasure to see the @ACSChemRev review on Gaussian process regression for chemistry & materials out in print! Read it #openaccess at
https://t.co/XS8Q4IEnIP

[Google DeepMind @GoogleDeepMind]

Today with @emblebi, we're launching the #AlphaFold Protein Structure Database, which offers the most complete and accurate picture of the human proteome, doubling humanity’s accumulated knowledge of high-accuracy human protein structures - for free: https://t.co/vtBGmTkKhy 1/

[Google AI @GoogleAI]

Today we present two studies that demonstrate how a quantum processor can be used to accurately simulate the properties of quantum materials — thin wires and interacting electron sheets — allowing accurate computation of their physical properties. https://t.co/gBDfUjndrc

[FAIR Chemistry @OpenCatalyst]

NeurIPS 2021: Open Catalyst Challenge!

Details: https://t.co/x3d9Tzc46y
Discussion Forum:https://t.co/lkh5qyPAuP
Leaderboard: https://t.co/zeF1XJu65e

[John Preskill @preskill]

I was asked by Tony Hey to contribute a chapter for a new edition of the Feynman Lectures on Computation. All I had to do (in about 40 pages) is explain what’s happened in quantum computing since Feynman first talked about it 40 years ago.

https://t.co/8LS4Voc0hK

[Jonathan Vandermause @jonpvandermause]

Thrilled to share our tutorial on the flare code! The notebook demos some nifty tools we've implemented recently, including sparse GP force fields with uncertainties and fast multi-element descriptors (+gradients) derived from the atomic cluster expansion #OpenSource #OpenScience

[Materials Intelligence Research @ Harvard @Materials_Intel]

Want to learn a many-body force field on the fly? Check out our tutorial on FLARE given by @jonpvandermause, now on YouTube! You can run the tutorial in Google Colab here: https://t.co/YmF7FdkCSK #ActiveLearning #BayesianForceFields

https://t.co/21Exe9D5Xe

[Jonathan Vandermause @jonpvandermause]

Excited to be a part of this very nice work by my colleague and collaborator @YuuuXie! Here Yu shows how you can accelerate GP-based ML force fields to perform uncertainty-aware molecular dynamics simulations at scale, going all the way up to millions of atoms.

[Yu Xie @YuuuXie]

So happy that our work with @Materials_Intel @bkoz37
on accelerated Bayesian active learning w/ mapped Gaussian process of 2+3 body kernel is now on npj Comput. Mater. @Nature_NPJ!

Check out the article:
https://t.co/XrnF937zir

#MaterialsScience #compchem #MachineLearning