Universal Representation for Code

Linfeng Liu, Hoan Nguyen, George Karypis +1

Learning from source code usually requires a large amount of labeled data. Despite the possible scarcity of labeled data, the trained model is highly task-specific and lacks transf…