Prospects of applying neural networks to supercompilation and equality saturation
Abstract:
Many methods of program transformation (including supercompilation and equality saturation) may be formulated as a set of term or graph rewriting rules that are applied in the order defined by heuristics. These heuristics are usually created by hand, so it is very interesting to automate their creation, e.g. with machine learning. In this paper we survey some approaches to tackle these problems with neural networks.
Keywords:
machine learning, neural networks, program analysis, program transformation
