Programming with “big code”

Abstract

The vast amount of code available on the web is increasing on a daily basis. Open-source hosting sites such as GitHub contain billions of lines of code. Community question-answering sites provide millions of code snippets with corresponding text and metadata. The amount of code available in executable binaries is even greater. In this talk, I will cover recent research trends on leveraging such “big code” for program analysis, program synthesis and reverse engineering. We will consider a range of semantic representations based on symbolic automata [11,15], tracelets [3], numerical abstractions [13,14], and textual descriptions [1,22], as well as different notions of code similarity based on these representations. To leverage these semantic representations, we will consider a number of prediction techniques, including statistical language models [19,20], variable order Markov models [2], and other distance-based and modelbased sequence classification techniques. Finally, I will show applications of these techniques including semantic code search in both source code and stripped binaries, code completion and reverse engineering.