BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Neural Equivariant Interatomic Potentials - Simon Batzner\, Harvar
 d University
DTSTART:20210308T163000Z
DTEND:20210308T173000Z
UID:TALK157732@talks.cam.ac.uk
CONTACT:Bingqing Cheng
DESCRIPTION:Representations of atomistic systems for machine learning must
  transform predictably under the geometric transformations of 3D space\, i
 n particular rotation\, translation\, mirrors\, and permutation of atoms o
 f the same species. These constraints are typically satisfied by means of 
 atomistic representations that depend on scalar distances and angles\, lea
 ving the representation invariant under the above transformations. Invaria
 nce\, however\, limits the expressivity and can lead to an incompleteness 
 of  representations. In order to overcome this shortcoming\, we recently i
 ntroduced Neural Equviariant Interatomic Potentials [1]\, a Graph Neural N
 etwork approach for learning interatomic potentials that uses a SE(3)-equi
 variant representation of atomic environments. While most current Graph Ne
 ural Network interatomic potentials use invariant convolutions over scalar
  features\, NequIP instead employs equivariant convolutions over geometric
  tensors (scalar\, vectors\, …)\, providing a more information-rich mess
 age passing scheme. \nIn my talk\, I will first motivate the choice of an 
 equivariant representation for atomistic systems and demonstrate how it al
 lows for the design of interatomic potentials at previously unattainable a
 ccuracy. I will discuss applications on a diverse set of molecules and mat
 erials\, including small organic molecules\, water in different phases\, a
  catalytic surface reaction\, glass formation of a lithium phosphate\, and
  Li diffusion in a superionic conductor. I will then show that NequIP can 
 predict structural and kinetic properties from molecular dynamics simulati
 ons in excellent agreement with ab-initio simulations. The talk will then 
 discuss the observation of a remarkable sample efficiency in equivariant i
 nteratomic potentials which outperform existing neural network potentials 
 with up to 1000x fewer training data and rival or even surpass the sample 
 efficiency of kernel methods. Finally\, I will discuss potential reasons f
 or the high sample efficiency of equivariant interatomic potentials.  \n\n
 [1] https://arxiv.org/abs/2101.03164
LOCATION:virtual ZOOM meeting ID: 263 591 6003\, Passcode: 000042\, https:
 //us02web.zoom.us/j/2635916003?pwd=ZlBEQnRENGwxNmJGMENGMWxjak5nUT09
END:VEVENT
END:VCALENDAR