Code Vectors: Understanding Programs Through Embedded Abstracted Symbolic Traces
This work addresses the challenge of making programs amenable to learning algorithms for researchers and practitioners in software engineering and AI, representing an incremental improvement in program representation methods.
The paper tackles the problem of representing programs for machine learning by using abstractions of symbolic execution traces to learn word embeddings, achieving 93% top-1 accuracy on a benchmark of over 19,000 API-usage analogies from the Linux kernel and showing that semantic abstractions provide nearly triple the accuracy of syntactic ones.
With the rise of machine learning, there is a great deal of interest in treating programs as data to be fed to learning algorithms. However, programs do not start off in a form that is immediately amenable to most off-the-shelf learning techniques. Instead, it is necessary to transform the program to a suitable representation before a learning technique can be applied. In this paper, we use abstractions of traces obtained from symbolic execution of a program as a representation for learning word embeddings. We trained a variety of word embeddings under hundreds of parameterizations, and evaluated each learned embedding on a suite of different tasks. In our evaluation, we obtain 93% top-1 accuracy on a benchmark consisting of over 19,000 API-usage analogies extracted from the Linux kernel. In addition, we show that embeddings learned from (mainly) semantic abstractions provide nearly triple the accuracy of those learned from (mainly) syntactic abstractions.