avatar

Peter Wirnsberger

DeepMind, London, UK

Work and Education

  • Now

    Senior Research Scientist | DeepMind

    My current research interests include a broad range of topics in statistical mechanics and ML, ranging from generative modelling for molecular simulation, learning large-scale simulators, ML-based free energy estimators, enhanced sampling techniques, and causal decision making.

  • 2018

    PhD in Theoretical Chemistry | University of Cambridge

    My PhD, under the supervision of Daan Frenkel, comprised statistical modelling and advanced simulation techniques to study the behaviour of polar solvents far from thermal equilibrium. I contributed to the molecular dynamics package LAMMPS and won an award for my PhD thesis.

  • 2014

    MSc in Computational Physics | University of Vienna

    This course was a deep dive into selected topics ranging from statistical mechanics to chaos theory, with extensive practical training in Monte Carlo and molecular dynamics (MD) simulations. My Master's project was jointly supervised by Christoph Dellago and Daan Frenkel.

  • 2012

    MPhil in Scientific Computing | University of Cambridge

    This course focused on numerical solutions to differential equations and High Performance Computing (C++, OpenMP, MPI and CUDA). For my dissertation, supervised by Nikos Nikiforakis and Franck Monmont, I implemented the C++ CFD code Impec2D to investigate the dependence of common oil reservoir solvers on the orientation of the grid.

  • 2011

    BSc in Physics | University of Vienna

    The 3-year undergraduate programme was an excellent training in maths, physics, mathematical modelling and scientific computing. My Bachelor's thesis was concerned with modelling atmospheric flows in Cartesian and spherical geometry.

  • 2008

    Social work | Red Cross

    As an alternative to the compulsory military service, I worked for the Red Cross for a period of 9 months and became a paramedic.

  • 2007

    Secondary school | software engineering

    The 5-year long technical high school HTL Villach provided me with a solid background in software engineering, in particular C/C++ and Java, algorithms, data structures and data bases (SQL, PL/SQL).

Teaching and Internships


Microsoft
Summer 2017
Machine Learning | Research Internship
In my 3-month internship at MS Research Cambridge I employed ML techniques to improve the user experience in chat-based communication systems.
Cambridge/ Vienna
2011-2017
Supervising undergraduate students | Teaching
Statistical Mechanics, Computer simulation methods in chemistry and physics, Scientific Computing, Introduction to Mathematics for Physicists.
University of Cambridge
2013
Computational Chemistry | Intern
During my Master's project, I worked for 6 months in the group of Daan Frenkel to study thermally induced ordering phenomena in polar liquids.
University of Cambridge
2011
High Performance Computing | Training
This 2-week course covered basic HPC tools for profiling and optimisation, parallel architectures as well as OpenMP and MPI.
University of Cambridge
2010
Scientific Computing | Summer intern
I ran CFD simulations with adaptive mesh refinement to study advection in the 'Cubed Sphere' geometry (supervised by Nikos Nikiforakis).
University of Vienna
2009
Computational Physics | Summer intern
I investigated dipole moment correlations of one-dimensional water chains confined inside carbon nanotubes in the group of Christoph Dellago.
Infineon Tech., Villach
2008
Production and Manufacturing | Work
I operated a scanning electron microscope for visual detection of manufacturing defects in silicon wavers (3 months).
German Bank, Eschborn
2005
Web programming | Summer intern
I helped to adapt the internal German Bank search engine to mobile devices using HTML and JSP.

Open source contributions


Github
2022
Normalizing flows for atomic solids
Source code, trained models, example colabs and a tutorial for our paper (link).
Github
2015-2017
LAMMPS
Additional functionalities to the Large-scale Atomic/Molecular Massively Parallel Simulator for my PhD research written in C++ (see fix_ehex.cpp and fix_rattle.cpp).
Github
2017
Impec2D
A Computational Fluid Dynamics code for my MPhil research written in C++. The parallalised software uses the HYPRE multigrid solver, LAPACK, BLAS, and Boost libraries (link).

Awards and fellowships


2019
Highly Commended Thesis Prize
This prize was awarded to me by the Department of Chemistry, University of Cambridge, in recognition of my PhD Thesis.
2017
Microsoft Azure Research Award
Microsoft supports my research with access to their cloud computing service Azure and computing time for an entire year.
2017
Microsoft Hackathon Winner
My team was one of the global winners of the Microsoft Hackathon 2017 with more than 18,000 participants worldwide.
2017
ESI Junior Research Fellowship
This fellowship awarded by the Austrian Erwin Schrödinger Institute (ESI) supports my research on thermally induced monopoles.
2014
Honorary Award of the Austrian Federal Ministry of Science, Research and Economy
This prestigious price was awarded to me by the Austrian Minister of Science, Research and Economy to honour my outstanding performance while studying at the University of Vienna.
2014
Member of the Dean's List
This faculty internal price was awarded to me to honour my performance as a student at the Faculty of Physics, University of Vienna.
2014
Doctoral Fellowship of the Austrian Academy of Sciences
The DOC Fellowship provides full funding for my PhD research.
2013
Excellence Grant of the Federation of Austrian Industry Carinthia
This grant supported my ERASMUS internship in Cambridge.
2011
Schlumberger Departmental Award
Schlumberger Cambridge Research kindly provided full funding for my MPhil course.

Selected publications


2022
MultiScale MeshGraphNets
M. Fortunato, T. Pfaff, P. Wirnsberger, A. Pritzel, P. Battaglia, ICML AI4Science Workshop arXiv
2022
Normalizing flows for atomic solids
P. Wirnsberger, G. Papamakarios, B. Ibarz et al., Mach. Learn.: Sci. Technol. doi
2021
Symetric: measuring the quality of learnt hamiltonian dynamics inferred from vision
I. Higgins, P. Wirnsberger, A. Jaegle and A Botev, NeurIPS link
2021
Which priors matter? Benchmarking models for learning latent dynamics
A. Botev, A. Jaegle, P. Wirnsberger, D. Hennes and I. Higgins, NeurIPS (Datasets and Benchmark) link
2020
Targeted free energy estimation via learned mappings
P. Wirnsberger, A.J. Ballard et al., J. Chem. Phys. doi
2018
Theoretical prediction of thermal polarisation
P. Wirnsberger, C. Dellago, and D. Frenkel and A. Reinhardt, Phys. Rev. Lett. doi
2017
Numerical evidence for thermally induced monopoles
P. Wirnsberger, D. Fijan, R. A. Lightwood, A. Šarić, C. Dellago, and D. Frenkel, Proc Natl Acad Sci USA. doi
2015
An enhanced version of the heat exchange algorithm with excellent energy conservation properties
P. Wirnsberger, D. Frenkel, and C. Dellago, J. Chem. Phys. doi

Contact

Email
peter.wirnsberger at gmail.com
Homepage
pw359.github.io
Github
pw359