Factory functions for creating mlr3torch-based deep learning models. Phase 2 scope: Multi-Layer Perceptron (MLP) only.
Details
Why MLP only for Phase 2: - Tabular omics data typically doesn't have spatial/temporal ordering - CNNs assume local structure (adjacent features correlated) - misleading for most omics - MLPs are the appropriate baseline for unordered feature vectors - Focus is on proper nested CV and leakage prevention, not architecture variety