classification of X-ray Diffraction Patterns Shreyaa Raghavan Tonio - - PowerPoint PPT Presentation

Shreyaa Raghavan Tonio Buonassisi, Zhe Liu MIT Photovoltaic Research Lab Contact: shreyaar@mit.edu

Tackling Data Scarcity in Materials Research: Using Semi-supervised, Adversarial Training to Improve classification of X-ray Diffraction Patterns

X-ray Diffraction (XRD) Pattern Classification

▪ A typical machine learning problem ▪ Classification of crystals by space groups and dimensionality ▪ Currently, uses experimental data & computer- generated, synthetic data during training

Figure 1. Examples of perovskite XRD patterns with different dimensionalities1 (i.e. 0D, 2D, 3D)

1Sun, S. et al., Joule 3, 1437–1451 (2019).

XRD Pattern Example

(1) Data Scarcity regarding generating

labeled experimental data1

(2) Simulated data in the training can be

detrimental to the classifier

Challenges with Current Classifier (autoXRD)

Goal: Mimic experimental data and improve efficacy of non- experimental data (with generative adversarial network – GAN2)

1Oviedo, F., Ren, Z., Sun, S. et al. npj Comput Mater 5, 60 (2019).

• 2A. Shrivastava, T. Pfister, O. Tuzel, J. Susskind, W. Wang, and R. Webb. Learning from simulated and unsupervised images through adversarial training.

arXiv:1612.07828, 2016.

Proposed Method 1: Using Generative Adversarial Training

Simulated XRD input Binary Classification Real vs Fake Refiner Model Refined XRD Example Unlabeled Experimental XRD example Discriminator Model Model Update Model Update

Main Advantage: Training with Unlabeled Experimental Data

Proposed Method 2: Gaussian Filter

Effect of Gaussian Filter on simulated XRD data (i.e. widening peaks) Effect of Refiner Model VS

Results

Augmented Data 500 Data Accuracy (%) 1000 Data Accuracy (%) 2000 Data Accuracy (%) 4000 Data Accuracy (%) Simulated 12.7 23.9 26.2 34.6 Refiner Model A (20 to 1) 39.8 51.2 53.2 62.9 Refiner Model B (30 to 1) 11.0 17.3 44.2 49.5 Gaussian Filter 11.8 21.6 38.2 49.8

Table 1. Accuracies after 5-Fold Cross Validation of Space Group Classification using Proposed Methods and no Experimental Data

▪ Accelerating characterization tasks with machine/deep learning ▪ Generalizable to tackle data scarcity in materials research and other

fields (where there’s lack of large, labeled dataset)

Future Work

References

[1] Christopher Bowles et al. Gan augmentation: Augmenting training data using generative adversarial networks. ArXiv, abs/1810.10863, 2018. [2] Oviedo, F., Ren, Z., Sun, S. et al. Fast and interpretable classification of small X-ray diffraction datasets using data augmentation and deep neural networks. npj Comput Mater 5, 60 (2019). https://doi.org/10.1038/s41524-019-0196-x [3] A. Shrivastava, T. Pfister, O. Tuzel, J. Susskind, W. Wang, and R. Webb. Learning from simulated and unsupervised images through adversarial training. arXiv preprint arXiv:1612.07828, 2016. [4] Sun, S. et al. Accelerated development of perovskite-inspired materials via high-throughput synthesis and machine-learning diagnosis. Joule3, 1437–1451 (2019). [5] https://github.com/mjdietzx/SimGAN

Thank You! ☺

Contact: shreyaar@mit.edu