Learning Solution Operators for Partial Differential Equations via Monte Carlo-Type Approximation