Fig. 1: Workflow to identify transition regimes in pattern-forming processes via self-supervised learning.
From: Inferring topological transitions in pattern-forming processes with self-supervised learning
a We simulate the dynamical evolution of the physical system for a broad range of process parameters. Next, we project the final state of the microstructural pattern into a latent space (using a pre-trained ResNet-50 v235). We regress on these latent dimensions to estimate the original process parameters. b To detect specific classes of microstructural patterns, we evaluate the model error by predicting the corresponding initial process parameters. By measuring the change in sensitivity of forming specific patterns for various input process parameters, we learn where the transition regime(s) might occur.
