DICA identifies latent components from nonlinear mixtures by maximizing Jacobian volume, bringing identifiability without source independence, Jacobian sparsity, or auxiliary signals.

Stability analysis reveals that appropriately controlled Jacobian's spectra lead to high inference performance