Multi-Fidelity Manifold Learning for Feature Encoding: A Hyperparameter-Driven Approach

Our proposed MF module maps high-dimensional feature-encoded information to three low-dimensional manifolds of different dimension sizes: n, n/2, and n/4. These dimension sizes act as hyperparameters, allowing for a flexible and adaptable approach to feature representation. The dimension size of the high-dimensional manifolds is set to match the dimension of the input feature-encoded information.