Voting Data-Driven Regression Learning for Accelerating Discovery of Advanced Functional Materials and Applications to Two-Dimensional Ferroelectric Materials

Abstract

Regression machine learning is widely applied to predict various materials. However, insufficient materials data usually leads to a poor performance. Here, we develop a new voting data-driven method that could generally improve the performance of regression learning model for accurately predicting properties of materials. We apply it to investigate a large family (2135) of two-dimensional hexagonal binary compounds focusing on ferroelectric properties and find that the performance of the model for electric polarization is indeed greatly improved, where 38 stable ferroelectrics with out-of-plane polarization including 31 metals and 7 semiconductors are screened out. By an unsupervised learning, actionable information such as how the number and orbital radius of valence electrons, ionic polarizability, and electronegativity of constituent atoms affect polarization was extracted. Our voting data-driven method not only reduces the size of materials data for constructing a reliable learning model but also enables to make precise predictions for targeted functional materials.

Predicting quantum potentials by deep neural network and metropolis sampling

“…( 5) are randomly obtained according to the probability in Eq. (7). Other parts including the NN are the same as the QPNN.…”

“…(2) and the positions {r n } sampled from the probability distribution P(r) in Eq. (7). With the loss L → 0, the NN would give a potential U θ (r n ) → V (r) satisfying the Schrödinger equation (note V (r) denotes the "correct" potential that we expect the NN to give).…”

“…On one hand, it provides alternative or more powerful tools to solve the problems that are challenging for the conventional approaches. For instance, neural network (NN), which is widely accepted as the most powerful ML model, is utilized to design functional materials with much higher efficiency than human experts [2][3][4][5][6][7]. One popular way is to apply to ML model to fit the relations between the experimental or numerical data and the target physical quantities.…”

“…For MPNN, we use the loss function given in Eq (7). where {r n } are obtained by Metropolis sampling.…”

High-efficient ab initio Bayesian active learning method and applications in prediction of two-dimensional functional materials

“…2D ferroelectric materials [20][21][22][23][24][25] have been intensively studied due to a wide range of promising applications in electromechanical transducer, 26 ferroelectric field effect transistor 27 and ferroelectric tunnel junctions, 28 etc. More importantly, as there is the depolarizing electrostatic field in 2D materials, the electrical polarization of 2D materials is generally low, 20,22 limiting their further applications.…”