A local-hybrid functional has been developed with a composite local mixing function that combines a neural network and a strong-correlation model. This development may be related to learning local and semi-local density functionals from exact exchange-correlation potentials and energies. The details and implications of this functional are being reported in scientific research.