Scaling up Self-Explanatory Simulators: Polynomial-Time Compilation

Abstract

Self-explanatory simulators have many potential applica-tions, including supporting engineering activities, intelligent tutoring systems, and computer-based training systems. To fully realize this potential requires improving the technology to efficiently generate highly optimized simulators. This paper describes an algorithm for compiling self-explanatory simulators that operates in polynomial time. It is capable of constructing self-explanatory simulators with thousands of parameters, which is an order of magnitude more complex than any previous technique. The algorithm is fully imple-mented, and we show evidence that suggests its performance is quadratic in the size of the system being simulated. We also analyze the tradeoffs between compilers and interpret-ers for self-explanatory simulation in terms of application-imposed constraints, and discuss plans for applications. 1.