Figure 2
From: Reduced order modeling for flow and transport problems with Barlow Twins self-supervised learning
Example 1—results: (a) mean squared error (MSE) of each model (please refer to Table 2), and blue texts represent a mean value of the box plots—here we show that BT–AE 16 gives performance similar to POD-based approaches, but AE and DC–AE models do not, (b) data compression loss for validation set (Eq. 18), (c) mapping using ANN loss for validation set (Eq. 19), (d) latent space plot of DC–AE 16 Q model, and (e) latent space plot of BT–AE 16 Q model. Latent space plots are constructed using t-Distributed Stochastic Neighbor Embedding (t-SNE). Different colors represent each value of \(\textrm{Ra}\) value. We calculate the t-SNE plots using Scikit-Learn package using its default setting and perplexity of 15.
