Announcement of a Tensorflow + Hoomd-blue plugin

[Andrew White]

Hi hoomd-users,

I'd like to announce that we have developed a hoomd-blue tensorflow plugin that allows you to integrate tensorflow models with hoomd-blue simulations both on the GPU. Example applications are computing collective variables, implementing free energy methods, learning neural network force fields, and using neural network force fields trained from other datasets (e.g., quantum). Tensorflow is written in Python, so this can be done in your usual hoomd-blue input scripts.

You can find the plugin along with documentation here: https://github.com/ur-whitelab/hoomd-tf#hoomd-tf and we have a pre-print of our publication describing the plugin here: https://chemrxiv.org/articles/A_GPU-Accelerated_Machine_Learning_Framework_for_Molecular_Simulation_Hoomd-Blue_with_TensorFlow/8019527/1. Examples of its use in the README include computing hydrogen bonding number, a radial distribution function, mapping to a coarse-grain model, and implementing the Lennard-Jones force field in hoomd-blue.

If you have any issues or requests for features, please use the github interface at https://github.com/ur-whitelab/hoomd-tf/issues

Thank you,

Andrew White

Asst Prof. of Chemical Engineering

University of Rochester

http://thewhitelab.org/