Table 1 The recent advances in DL-driven crystallographic symmetry classification and phase identification

From: Deep learning for symmetry classification using sparse 3D electron density data for inorganic compounds

  

Perturbation Incorporation

Data size

Problem size

 

Test accuracy (%)

Authors

Data Generalizability

Preferred orientation, peak shifting, peak broadening, noise, etc

# of XRD patterns

# of classes

Diffraction Type

Top one Accuracy

Park et al.13

Generalizable

Fair

150,000

7/101/230

Conventional 1D

94/84/81

Ziletti et al.40

Narrow (eight elemental solids)

N/A

100,000

7

Virtual 2D

100

Oviedo et al.14

Narrow (thin-film metahalides)

Good

2000

7

Conventional 1D

99.1

Vecsei et al.15

Generalizable

Fair

128,404

7/230

Conventional 1D

85/76

Lee et al.22

Narrow (Li-Sr-Al-O)

Fair

188,000

9000

Conventional 1D

99

Suzuki et al.16

Generalizable

Fair

199,391

7/230

Conventional 1D

92/80

Maffettone et al.27

Narrow (Ni-Co-Al)

Good

600,000

6

Conventional 1D

94.7

Tiong et al.42

Pseudo- Generalizable

N/A

108,658

72

Virtual 2D

80.12

Aguiar et al.43

Generalizable

N/A

571,340

7

Azimuthal integration profile 1D

85.87

Szymanski et al.26

Narrow (Li-Mn-Ti-O-F)

Good

21,000

140

Conventional 1D

94

Wang et al.28

Narrow (Metal-Organic Frameworks)

Good

72,864

1012

Conventional 1D

N/A 96.7 (Top-5 Acc.)

Massuyeau et al.24

Narrow (Perovskite & non-Perovskite)

N/A

998

2

Conventional 1D

>85

Lee et al.20

Generalizable

N/A

189,476

7/101/230

Conventional 1D

92/81/79

Lee et al.21

Generalizable

Good

1,974,760

7/101/230

Conventional 1D

93/87/84

Schuetzke et al.18

Narrow (Iron ore & cement)

Good

100,000~500,000

28/76

Conventional 1D

95/99

Schuetzke et al.19

Pseudo- Generalizable

Good

30,000

500

Conventional 1D

98.9

Salgado et al.44

Generalizable

Good (No perturbations)

1,200,000 (171,006)

7/230

Conventional 1D

86/77 (96/94)

  1. This table outlines key attributes of various DL approaches, encompassing data generalizability, perturbation incorporation, data size, problem size, and diffraction type. ‘Data generalizability’ refers to the diversity range of inorganic material systems used, from narrow to broad. Datasets focused on specific chemical compositions are categorized as having a ‘narrow’ range. The ‘perturbation incorporation’ column evaluates the variety of perturbations applied in data simulation to assess the method’s robustness. ‘Data size’ and ‘problem size’ columns denote the number of data points and labels utilized for training the DL models. The ‘diffraction type’ column clarifies the fundamental nature of the data employed. Lastly, the ‘test accuracy’ column displays the outcome accuracies as top-one accuracy.
Back to article page