Neural networks trained for inverse problems provide deterministic convergence and recovery guarantees.
