Physics-Driven Learning for Inverse Problems in Quantum Chromodynamics