Table 1 Prediction of mobility parameters in the case of the spinodal decomposition problem.

From: Inferring topological transitions in pattern-forming processes with self-supervised learning

View full size image

  

Phase A

Phase B

Instance index

Phase fraction

Predicted mobility

Target mobility

Prediction error (Δi)

Sensitivity score (\({{{{\mathcal{S}}}}}_{i}\))

Predicted mobility

Target mobility

Prediction error (Δi)

Sensitivity score (\({{{{\mathcal{S}}}}}_{i}\))

1042

0.34766

0.3758

0.9408

0.5650

1.7700

0.4418

0.5062

0.0644

15.5344

2106

0.34964

0.3671

0.1338

0.2332

4.2874

0.2488

0.3725

0.1237

8.0867

4157

0.35344

0.4290

0.3608

0.0682

14.6568

0.1188

0.0979

0.0209

47.8391

2868

0.49802

0.3062

0.8484

0.5422

1.8443

0.3434

0.2021

0.1413

7.0785

4496

0.50287

0.2445

0.0432

0.2013

4.9672

0.3609

0.7487

0.3878

2.5784

4545

0.50289

0.1002

0.0464

0.0538

18.6006

0.2371

0.2705

0.0333

29.9854

814

0.64599

0.4386

0.3852

0.0533

18.7477

0.4604

0.3713

0.0891

11.2179

1543

0.64969

0.2343

0.2529

0.0186

53.6827

0.4152

0.3636

0.0516

19.3968

4751

0.65144

0.4143

0.5127

0.0984

10.1611

0.4619

0.8399

0.3780

2.6454

  1. An example of instances with high and low sensitivity scores. For the microstructure insets, yellow denotes the A phase and purple the B phase. The error Δi is defined as \({\Delta }_{i}=| \hat{y}-y|\), where \(\hat{y}\) is the prediction and y the true value. The sensitivity score is defined as \({{{{\mathcal{S}}}}}_{i}={\Delta }_{i}^{-1}\). Instances 1042, 2106, 4157 belong to regime A as identified in Fig. 4b. Instances 2868, 4496, 4545 belong to regime B as identified in Fig. 4b. Instances 814, 1543, 4751 belong to regime C as identified in Fig. 4b. Note that the high sensitivity scores (in bold) correspond to different input process parameters depending on the regime of importance. Each regime is separated by horizontal lines.
Back to article page