Self-Assembly is an autopoietic phenomenon in which complex structures ‘emerge’ from a system of autonomous entities without a master plan. Research into such sys- tems often faces the forward, backward and yield problems. These are usually solved through simulations and the use of extremely complicated heuristics. This research explores the plausibility of using modern machine learning techniques to achieve a func- tional approximation of a sophisticated Kinetic Monte Carlo self-assembly simulator in order to produce a data-driven solution to the forwards and backwards problem.