Discussion of: An Automatic Framework to Build Neural Network-based Surrogate for High-Performance Computing Applications

David Notes

• Use of "Customized Autoencoder" to sparsify inputs is interesting.
  • Are `const` parameters the primary values being filtered?
• User inputs place boundaries on accuracy and speed when optimizing topology and input features.
  • How is Bayesian used? (what are priors?)
• Feature to introduce perturbation into input features in cases where insufficient variation of data are available.(section 3.1, Step 3)

Cissie

• Slides: Auto-HPCNet vs PHASM
• Related Github folder by the 1st author: https://github.com/wdong5/AutoHPC-autoencoder