Machine learning for high entropy alloys

High entropy alloys are an exciting class of new materials. Even though they often combine 3, 4, 5 or more different principal elemental components in near equiatomic ratios, they somehow avoid the expected formation of intermetallics. Instead, they often have a single or perhaps a couple of simple FCC, BCC, HCP or other phases with remarkablly extended solid-solubility that traditional Hume Rothery rules would never have suggested! Check out my previous video on HEA basics if you want to learn more!
This video provides an overview of machine learning approaches meant to predict phase composition and properties of these alloys. I summarize what we can learn (good and bad!) from a dozen or so publications and also discuss HEA databases, best practices, interpretability, and some truly novel approaches that some researchers have utilized.    • What are high entropy alloys?  

links to videos discussed:
best practices    • 17. Best Practices in Materials Infor...  
bootstrapping    • 18. Ensemble techniques  
splitting data    • 16. Splitting data in train/val/test ...  
feature weights    • 18. Ensemble techniques  
metrics in ML    • 15. Metrics and evaluation of machine...  
GANs    • Guest Lecture (Ryan Murdock of Adobe)...  

0:00 why care about phase predictions in HEAs
2:14 phase prediction paper 1
5:22 features, Hume-Rothery rules
7:25 accuracy vs loss vs per class performance
10:12 phase prediction paper 2
12:48 phase prediction paper 3
16:12 phase prediction paper 4
18:29 genetic algorithm feature selection
21:20 phase prediction paper 5
24:31 GAN for data augmentation
28:55 phase prediction paper 6
30:37 takeaways from phase prediction
33:19 property prediction paper 1
37:40 property prediction paper 2
41:15 property prediction paper 3
43:27 property prediction paper 4
47:11 property prediction paper 5
49:50 property prediction paper 6
51:35 clever paper using VAE for order parameter
57:20 interpretability
1:01:01 data sets and active learning