Deep learning framework for material design space exploration using active transfer learning and data augmentation

Kim, Yongtae; Kim, Youngsoo; Yang, Charles; Park, Kundo; Gu, Grace X.; Ryu, Seunghwa

doi:10.1038/s41524-021-00609-2

Cited by 101 publications

(59 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, this improvement seems to be specific to our test data as predictions on the relaxed validation data were better when the model was trained only on relaxed structures(Figure S2). The CGCNN-HD trained on only relaxed structures under predicted the structures in H-Relaxed likely because the training data contains relatively few transition metal hydrides and the bounded hyperbolic activation function, utilized in the CGCNN-HD's convolutional layers, has poor predictive power on unseen domains [48]. This poor predictive power on unseen domains is also the reason the CGCNN-HD models make poor predictions on the the high and low formation energy structures of the MP data.…”

Section: Resultsmentioning

confidence: 99%

Data-Augmentation for Graph Neural Network Learning of the Relaxed Energies of Unrelaxed Structures

Gibson¹,

Hire²,

Hennig³

2022

Preprint

View full text Add to dashboard Cite

Computational materials discovery has continually grown in utility over the past decade due to advances in computing power and crystal structure prediction algorithms (CSPA). However, the computational cost of the ab initio calculations required by CSPA limits its utility to small unit cells, reducing the compositional and structural space the algorithms can explore. Past studies have bypassed many unneeded ab initio calculations by utilizing machine learning methods to predict formation energy and determine the stability of a material. Specifically, graph neural networks display high fidelity in predicting formation energy. Traditionally graph neural networks are trained on large data sets of relaxed structures. Unfortunately, the geometries of unrelaxed candidate structures produced by CSPA often deviate from the relaxed state, which leads to poor predictions hindering the model's ability to filter energetically unfavorable prior to ab initio evaluation. This work shows that the prediction error on relaxed structures reduces as training progresses, while the prediction error on unrelaxed structures increases, suggesting an inverse correlation between relaxed and unrelaxed structure prediction accuracy. To remedy this behavior, we propose a simple, physically motivated, computationally cheap perturbation technique that augments training data to improve predictions on unrelaxed structures dramatically. On our test set consisting of 623 Nb-Sr-H hydride structures, we found that training a crystal graph convolutional neural networks, utilizing our augmentation method, reduced the MAE of formation energy prediction by 66% compared to training with only relaxed structures. We then show how this error reduction can accelerates CSPA by improving the model's ability to filter out energetically unfavorable structures accurately.

show abstract

Section: Resultsmentioning

confidence: 99%

Data-Augmentation for Graph Neural Network Learning of the Relaxed Energies of Unrelaxed Structures

Gibson¹,

Hire²,

Hennig³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…• The toolkit we developed provides a faster and more flexible structure-properties platform, which expedites the particlebased simulation and materials design procedures. • CuLSM opens the venue for high-throughput and highfidelity data generation to meet the increasing need for machine learning aided materials design protocols (Kim et al, 2021;Sui et al, 2021). • The work largely reduces the computational cost for predicting elasticity and fracture behaviors of complex materials systems, further accelerating the design phase through offering predictive insights for additive manufacturing and mechanical experimentation.…”

Section: Discussionmentioning

confidence: 99%

ImageMech: From Image to Particle Spring Network for Mechanical Characterization

2022

View full text Add to dashboard Cite

The emerging demand for advanced structural and biological materials calls for novel modeling tools that can rapidly yield high-fidelity estimation on materials properties in design cycles. Lattice spring model , a coarse-grained particle spring network, has gained attention in recent years for predicting the mechanical properties and giving insights into the fracture mechanism with high reproducibility and generalizability. However, to simulate the materials in sufficient detail for guaranteed numerical stability and convergence, most of the time a large number of particles are needed, greatly diminishing the potential for high-throughput computation and therewith data generation for machine learning frameworks. Here, we implement CuLSM, a GPU-accelerated compute unified device architecture C++ code realizing parallelism over the spring list instead of the commonly used spatial decomposition, which requires intermittent updates on the particle neighbor list. Along with the image-to-particle conversion tool Img2Particle, our toolkit offers a fast and flexible platform to characterize the elastic and fracture behaviors of materials, expediting the design process between additive manufacturing and computer-aided design. With the growing demand for new lightweight, adaptable, and multi-functional materials and structures, such tailored and optimized modeling platform has profound impacts, enabling faster exploration in design spaces, better quality control for 3D printing by digital twin techniques, and larger data generation pipelines for image-based generative machine learning models.

show abstract

“…In addition, the ML model must consider adequate design space between training set and test set. For example, if a variable with a higher concentration range than the training set used for model development is input into the developed ML model, the calibration performance has the possibility of deterioration [15]. In other words, the training set must contain a wide enough range of concentration of PM 2.5 .…”

Section: Dataset For Calibration Machine Learning (Ml) Modelingmentioning

confidence: 99%

Assessment and Calibration of a Low-Cost PM2.5 Sensor Using Machine Learning (HybridLSTM Neural Network): Feasibility Study to Build an Air Quality Monitoring System

et al. 2021

View full text Add to dashboard Cite

Commercially available low-cost air quality sensors have low accuracy. The improved accuracy of low-cost PM2.5 sensors allows the use of low-cost sensor systems to reasonably investigate PM2.5 emissions from industrial activities or to accurately estimate individual exposure to PM2.5. In this work, we developed a new PM2.5 calibration model (HybridLSTM) by combining a deep neural network (DNN) optimized in calibration problems and a long short-term memory (LSTM) neural network optimized in time-dependent characteristics to improve the performance of conventional calibration algorithms of low-cost PM sensors. The PM2.5 concentrations, temperature and humidity by low-cost sensors and gravimetric-based PM2.5 measuring instrument were sampled for a sufficiently long time. The proposed model was compared with benchmarks (multiple linear regression model (MLR), DNN model) and low-cost sensor results. The gravimetric measurements were used as reference data to evaluate sensor accuracy. For root-mean-square error (RMSE) for PM2.5 concentrations, the proposed model reduced 41–60% of error when compared with the raw data of low-cost sensors, reduced 30–51% of error when compared with the MLR model and reduced 8–40% of error when compared with the MLR model. R2 of HybridLSTM, DNN, MLR and raw data were 93, 90, 80 and 59%, respectively. HybridLSTM showed the state-of-the-art calibration performance for a low-cost PM sensor. In other words, the proposed ML model has state-of-the-art calibration performance among the tested calibration algorithms.

show abstract

Deep learning framework for material design space exploration using active transfer learning and data augmentation

Cited by 101 publications

References 54 publications

Data-Augmentation for Graph Neural Network Learning of the Relaxed Energies of Unrelaxed Structures

Data-Augmentation for Graph Neural Network Learning of the Relaxed Energies of Unrelaxed Structures

ImageMech: From Image to Particle Spring Network for Mechanical Characterization

Assessment and Calibration of a Low-Cost PM2.5 Sensor Using Machine Learning (HybridLSTM Neural Network): Feasibility Study to Build an Air Quality Monitoring System

Contact Info

Product

Resources

About