Research Data and Code for "Feasibility Mapping of Latent Representations Reveals Transferability Limits in Machine-Learning Models for Electrochemical Energy Storage Materials"

This release provides the complete codebase and curated datasets supporting our feasibility-mapping framework for interpreting latent representations in multitask machine-learning models for electrochemical energy-storage materials. The repository includes preprocessing scripts, ANN training, latent-space visualization (PCA/UMAP), domain divergence metrics (centroid distance, MMD²), linear-probe separability analysis, and supplementary diagnostics for porous carbons and MOFs.