Efficient sampling for decision making in materials discovery*

Abstract

Accelerating materials discovery crucially relies on strategies that efficiently sample the search space to label a pool of unlabeled data. This is important if the available labeled data sets are relatively small compared to the unlabeled data pool. Active learning with efficient sampling methods provides the means to guide the decision making to minimize the number of experiments or iterations required to find targeted properties. We review here different sampling strategies and show how they are utilized within an active learning loop in materials science.

Emerging Trends in Machine Learning: A Polymer Perspective

“…There have been many nonpolymer studies on the topics of autonomous formulation exploration and phase-mapping and many of them make use of theory informed or constrained models, similar to what was discussed in the Domain Knowledge section above. ,− In order to increase the accuracy of their phase-identification from X-ray diffraction measurements (XRD), Suram et al used a customized non-negative matrix factorization (NMF) approach in which they incorporated physical knowledge of solid state phase diagrams such as Gibb’s phase rule and XRD peak-shifting due to alloying. , Under similar motivations, Chen et al used an unsupervised, autoencoder approach in which they construct a latent subspace of meaningful variables and then express constraints with these variables . Kusne et al also leveraged domain knowledge in their agent but, interestingly, also demonstrated that employing multitask learning to combine the task of property optimization with that of identifying phase boundaries is more efficient than performing either task alone. , Finally, McDannald et al identify the magnetic ordering transition using neutron diffraction by encoding physical details of the measurement (e.g., hysteresis, appropriate parameter distributions) and further by automatically selecting from a set of analytical models for the final analysis .…”

An active machine learning approach for optimal design of magnesium alloys using Bayesian optimisation

“…One promising approach for expediting the discovery of metallic alloys with target mechanical properties is by using machine learning 5 – 8 . Machine learning (ML) accelerates new materials discovery by reducing the time and cost required for traditional trial-and-error approaches 6 , 9 , 10 , and utilising large datasets, advanced algorithms, and computational methods. This enables the acceleration of optimal alloy identification.…”

“…In particular, so-called active machine learning approaches, which combine human expertise with iterative model refinement, have demonstrated great potential in reducing the experimental burden and maximising the search efficiency in materials design 11 – 14 . Bayesian optimisation and adaptive design are methods following an active ML strategy, which require goal-directed iterative feedback 6 , 15 , 16 .…”