Abstract: Highlights•We exploit multi-spaces for disentangled representation learning.•We use Ricci curvature for local geometric-topological information.•We employ interactive learning for muti-space information exchange.•We use adaptive fusion to consider feature differences in diverse spaces.
Loading