Active Learning and Uncertainty Estimation

Shapeev, Alexander V.; Gubaev, Konstantin; Tsymbalov, Evgenii; Podryabinkin, Evgeny V.

doi:10.1007/978-3-030-40245-7_15

Cited by 19 publications

(17 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While performing ab initio calculations in regular intervals will discover all deviations of the model eventually, this variant of on-the-fly ML does not exploit any information about the already collected reference set and may thus lead to many redundant data points. More detailed reviews on uncertainty estimation and active sampling of PESs can be found in refs ( 226 ) and ( 227 ).…”

Section: Best Practices and Pitfallsmentioning

confidence: 99%

Machine Learning Force Fields

et al. 2021

View full text Add to dashboard Cite

In recent years, the use of machine learning (ML) in computational chemistry has enabled numerous advances previously out of reach due to the computational complexity of traditional electronic-structure methods. One of the most promising applications is the construction of ML-based force fields (FFs), with the aim to narrow the gap between the accuracy of ab initio methods and the efficiency of classical FFs. The key idea is to learn the statistical relation between chemical structure and potential energy without relying on a preconceived notion of fixed chemical bonds or knowledge about the relevant interactions. Such universal ML approximations are in principle only limited by the quality and quantity of the reference data used to train them. This review gives an overview of applications of ML-FFs and the chemical insights that can be obtained from them. The core concepts underlying ML-FFs are described in detail, and a step-by-step guide for constructing and testing them from scratch is given. The text concludes with a discussion of the challenges that remain to be overcome by the next generation of ML-FFs.

show abstract

Section: Best Practices and Pitfallsmentioning

confidence: 99%

Machine Learning Force Fields

et al. 2021

View full text Add to dashboard Cite

show abstract

“…According to Eqs. ( 26), (27), and (28), the total ML uncertainty σ on DOS(E) T derives from both the uncertainty on individual DOS predictions, σ a and the uncertainty on the phase space sampling associated with the committee of MLPs driving the dynamics, σ V .…”

Section: Finite-temperature Density Of Statesmentioning

confidence: 99%

“…The uncertainty is estimated on the basis of the spread of the predictions of an ensemble (committee) of independently trained ML models, e.g. by subsampling of the full training dataset [25][26][27][28][29] . These uncertainty quantification schemes provide qualitative information on the reliability of the ML predictions, and are widely used in the context of online and offline active learning, to identify regions of configuration space that need to be added to the training set 7,[30][31][32][33][34][35] .…”

Section: Introductionmentioning

confidence: 99%

“…by subsampling of the full training dataset [25][26][27][28][29] . These uncertainty quantification schemes provide qualitative information on the reliability of the ML predictions, and are widely used in the context of online and offline active learning, to identify regions of configuration space that need to be added to the training set 7,[30][31][32][33][34][35] . When appropriately calibrated 36 , committee models can further provide a quantitative assessment of the uncertainty in the prediction of an ML model, which can be readily propagated to estimate the error in properties that are obtained indirectly from the ML predictions such as vibrational spectra 37 .…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Uncertainty estimation for molecular dynamics and sampling

Imbalzano,

Zhuang,

Kapil

et al. 2020

Preprint

View full text Add to dashboard Cite

“…Even when uncertainty estimates are available to distinguish informative from uninformative inputs, ML potentials rely on atomistic simulations to generate new trial configurations and bootstrapping a potential becomes an infinite regress problem: the training data for the potential needs to represent the full PES, but thoroughly sampling the PES requires exhaustive sampling, which needs long simulations with a stable accurate potential. It is common to perform molecular dynamics (MD) simulations with NNbased models to expand their training set in an active learning (AL) loop 20,25,38 . MD simulations explore the phase space based on the thermodynamic probability of the PES.…”

mentioning

confidence: 99%

Differentiable sampling of molecular geometries with uncertainty-based adversarial attacks

2021

View full text Add to dashboard Cite

Neural network (NN) interatomic potentials provide fast prediction of potential energy surfaces, closely matching the accuracy of the electronic structure methods used to produce the training data. However, NN predictions are only reliable within well-learned training domains, and show volatile behavior when extrapolating. Uncertainty quantification methods can flag atomic configurations for which prediction confidence is low, but arriving at such uncertain regions requires expensive sampling of the NN phase space, often using atomistic simulations. Here, we exploit automatic differentiation to drive atomistic systems towards high-likelihood, high-uncertainty configurations without the need for molecular dynamics simulations. By performing adversarial attacks on an uncertainty metric, informative geometries that expand the training domain of NNs are sampled. When combined with an active learning loop, this approach bootstraps and improves NN potentials while decreasing the number of calls to the ground truth method. This efficiency is demonstrated on sampling of kinetic barriers, collective variables in molecules, and supramolecular chemistry in zeolite-molecule interactions, and can be extended to any NN potential architecture and materials system.

show abstract

Active Learning and Uncertainty Estimation

Cited by 19 publications

References 22 publications

Machine Learning Force Fields

Machine Learning Force Fields

Uncertainty estimation for molecular dynamics and sampling

Differentiable sampling of molecular geometries with uncertainty-based adversarial attacks

Contact Info

Product

Resources

About