Concorde improves link prediction via decoupled energy minimization

Concorde is an energy-based framework for link prediction in sparse graphs that decomposes likelihood into spectral, mixed-curvature, and continuous vector field branches.