Colmena: Scalable Machine-Learning-Based Steering of Ensemble Simulations for High Performance Computing

Abstract

Scientific applications that involve simulation ensembles can be accelerated greatly by using experiment design methods to select the best simulations to perform. Methods that use machine learning (ML) to create proxy models of simulations show particular promise for guiding ensembles but are challenging to deploy because of the need to coordinate dynamic mixes of simulation and learning tasks. We present Colmena, an open-source Python framework that allows users to steer campaigns by providing just the implementations of individual tasks plus the logic used to choose which tasks to execute when. Colmena handles task dispatch, results collation, ML model invocation, and ML model (re)training, using Parsl to execute tasks on HPC systems. We describe the design of Colmena and illustrate its capabilities by applying it to electrolyte design, where it both scales to 65536 CPUs and accelerates the discovery rate for high-performance molecules by a factor of 100 over unguided searches.

arXiv commentcamera-ready version for ML in HPC Environments 2021

References in corpus (8)

• Neural Message Passing for Quantum Chemistry
• Tune: A Research Platform for Distributed Model Selection and Training
• RLlib: Abstractions for Distributed Reinforcement Learning
• Parsl: Pervasive Parallel Programming in Python
• funcX: A Federated Function Serving Fabric for Science
• A Data Ecosystem to Support Machine Learning in Materials Science
• RADICAL-Cybertools: Middleware Building Blocks for Scalable Science
• Using Machine Learning at Scale in HPC Simulations with SmartSim: An Application to Ocean Climate Modeling

Cited by in corpus (5)

• funcX: Federated Function as a Service for Science
• UniFaaS: Programming across Distributed Cyberinfrastructure with Federated Function Serving
• Portable, heterogeneous ensemble workflows at scale using libEnsemble
• An Active Learning-Based Streaming Pipeline for Reduced Data Training of Structure Finding Models in Neutron Diffractometry
• High Throughput Training of Deep Surrogates from Large Ensemble Runs