Machine-learning-based dynamic-importance sampling for adaptive multiscale simulations

Bhatia, Harsh; Carpenter, Timothy S.; Ingólfsson, Helgi I.; Dharuman, Gautham; Karande, Piyush; Liu, Shusen; Oppelstrup, Tomas; Neale, Chris; Lightstone, Felice C.; Essen, Brian Van; Glosli, James N.; Bremer, Peer Timo

doi:10.1038/s42256-021-00327-w

Cited by 25 publications

(37 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Of special relevance to our work is MuMMI [24,37], which provides a bidirectional coupling of two scales -a macro and a micro -using ML for forward coupling and in situ feedback for backward. By using ML to dynamically select the most novel macro configurations [12], MuMMI continuously steers the multiscale simulation towards new exploration and, given enough time, can simulate every type of configuration either directly or as a proxy to a similar enough configuration. MuMMI also analyzes the ongoing micro scale simulations and can use their results to update the less-accurate, macro model, thereby, creating a self-healing mechanism, which, given enough time, will improve the accuracy of the coarser model.…”

Section: Related Workmentioning

confidence: 99%

“…The 30 nm × 30 nm patches extracted from the continuum data are evaluated for "novelty" in a reduced, 9-D representation generated by a metric learning approach implemented using a deep neural network. Similar to the work of Bhatia et al [12], a farthest-point sampling approach is used to identify novel configurations, although our patches are almost 55× larger (sampled on a 37×37 grid instead of 5×5). In the case of relevant CG frames, the conformational state of the RAS-RAF complex is coded using a 3-D representation, which unlike the representation of patches is not conducive to farthest-point sampling.…”

Section: The Three-scale Mummimentioning

confidence: 99%

“…Task 2: Select important patches/frames for spawning new CG/AA simulations. A custom, abstract API was developed using the DynIm framework [12,14] that was extended by both the Patch Selector and the (CG) Frame Selector to implement specific procedures. Both selectors operate on DynIm's high-dimensional point objects and, hence, are agnostic to the specific encoding of patches and frames.…”

Section: Workflow Managementmentioning

confidence: 99%

“…In particular, the recent presentation of MuMMI [37] offers a new paradigm for large multiscale simulations that couples two scalesa macro and a micro -using a dynamic sampling driven by machine learning (ML) [12] and in situ feedback. The MuMMI workflow [24] is capable of orchestrating such massive multiscale campaigns at unprecedented computational scales, with effective utilization of all of the heterogeneous resources on some of the largest machines on the planet (demonstrated on Sierra, using 176,000 CPU cores and 16,000 GPUs).…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Generalizable coordination of large multiscale workflows

Bhatia

Natale

Moon

et al. 2021

Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis

Self Cite

View full text Add to dashboard Cite

The advancement of machine learning techniques and the heterogeneous architectures of most current supercomputers are propelling the demand for large multiscale simulations that can automatically and autonomously couple diverse components and map them to relevant resources to solve complex problems at multiple scales. Nevertheless, despite the recent progress in workflow technologies, current capabilities are limited to coupling two scales. In the first-ever demonstration of using three scales of resolution, we present a scalable and generalizable framework that couples pairs of models using machine learning and in situ feedback. We expand upon the massively parallel Multiscale Machine-Learned Modeling Infrastructure (MuMMI), a recent,

show abstract

Section: Related Workmentioning

confidence: 99%

Section: The Three-scale Mummimentioning

confidence: 99%

Section: Workflow Managementmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Generalizable coordination of large multiscale workflows

Bhatia

Natale

Moon

et al. 2021

Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis

Self Cite

View full text Add to dashboard Cite

show abstract

“…proposed a data-driven method that trains deep neural networks to learn coarse-scale partial differential operators based on fine-scale data. In [8], Bhatia et al presented a novel paradigm of multiscale modeling that couples models at different scales using a dynamic-important sampling approach. A machine learning model is employed to dynamically sample in the phase space, hence enabling an automatic feedback from micro to macro scale.…”

Section: Introductionmentioning

confidence: 99%

Interfacing Finite Elements with Deep Neural Operators for Fast Multiscale Modeling of Mechanics Problems

Yin¹,

Zhang²,

Yu³

et al. 2022

Preprint

View full text Add to dashboard Cite

Multiscale modeling is an effective approach for investigating multiphysics systems with largely disparate size features, where models with different resolutions or heterogeneous descriptions are coupled together for predicting the system's response. The solver with lower fidelity (coarse) is responsible for simulating domains with homogeneous features, whereas the expensive high-fidelity (fine) model describes microscopic features with refined discretization, often making the overall cost prohibitively high, especially for timedependent problems. In this work, we explore the idea of multiscale modeling with machine learning and employ DeepONet, a neural operator, as an efficient surrogate of the expensive solver. DeepONet is trained offline using data acquired from the fine solver for learning the underlying and possibly unknown finescale dynamics. It is then coupled with standard PDE solvers for predicting the multiscale systems with new boundary/initial conditions in the coupling stage. The proposed framework significantly reduces the computational cost of multiscale simulations since the DeepONet inference cost is negligible, facilitating readily the incorporation of a plurality of interface conditions and coupling schemes. We present various benchmarks to assess accuracy and speedup, and in particular we develop a coupling algorithm for a timedependent problem, and we also demonstrate coupling of a continuum model (finite element methods, FEM) with a neural operator representation of a particle system (Smoothed Particle Hydrodynamics, SPH) for a uniaxial tension problem with hyperelastic material. What makes this approach unique is that a well-trained over-parametrized DeepONet can generalize well and make predictions at a negligible cost.

show abstract

Two decades of Martini: Better beads, broader scope

Marrink

Monticelli

Melo

et al. 2022

WIREs Comput Mol Sci

135

109

View full text Add to dashboard Cite

The Martini model, a coarse-grained force field for molecular dynamics simulations, has been around for nearly two decades. Originally developed for lipidbased systems by the groups of Marrink and Tieleman, the Martini model has over the years been extended as a community effort to the current level of a general-purpose force field. Apart from the obvious benefit of a reduction in computational cost, the popularity of the model is largely due to the systematic yet intuitive building-block approach that underlies the model, as well as the open nature of the development and its continuous validation. The easy implementation in the widely used Gromacs software suite has also been instrumental. Since its conception in 2002, the Martini model underwent a gradual refinement of the bead interactions and a widening scope of applications. In this review, we look back at this development, culminating with the release of the Martini 3 version in 2021. The power of the model is illustrated with key examples of recent important findings in biological and material sciences enabled with Martini, as well as examples from areas where coarse-grained resolution is essential, namely highthroughput applications, systems with large complexity, and simulations approaching the scale of whole cells.

show abstract

Machine-learning-based dynamic-importance sampling for adaptive multiscale simulations

Cited by 25 publications

References 24 publications

Generalizable coordination of large multiscale workflows

Generalizable coordination of large multiscale workflows

Interfacing Finite Elements with Deep Neural Operators for Fast Multiscale Modeling of Mechanics Problems

Two decades of Martini: Better beads, broader scope

Contact Info

Product

Resources

About