POPL'19論文との違い (査読者B)

[masuhar]

The following recent paper proposes a
similar technique for representing AST.
Uri Alon, Meital Zilberstein, Omer Levy, and Eran Yahav. 2019.
code2vec: learning distributed representations of code. Proc. ACM
Program. Lang. 3, POPL, Article 40 (January 2019), 29 pages. DOI:
https://doi.org/10.1145/3290353
Please add discussions comparing ASTToken2Vec and this study for
differences, potential issues, and so on.

[masuhar]

3290353.pdf

[masuhar]

• An embedding method for program text.
• The domain of embedding is a code snippet, a few lines of code. (Ours: one lexical token of code.)
• The applications are semantic labeling of code snippets. For example, a NN receives a nameless function definition of a few lines of code and answers "reverse array" as the function of the code. (Ours: input a partial input program, output a next few tokens.)
• It finds that a human-given identifier often can be decomposed into vector components of fragments of meanings. (equalsIgnoreCase = equals + toLowerCase) This fact justifies our use of word2vec like embedding for token prediction.
• It uses an attention model called "path-based" one, which can relate one AST node with the other one in a position like "go up tree by the name field, down the if statement, down the return statement."