TLDR: Perturb the code, check LLM robustness.

TLDR: Teach an LLM to fix code through (useful) rationales about the code and its execution.

TLDR: Check if a model can perform a roundtrip. If it does, that's good.

TLDR: Distill model predictions into interpretable patterns that can be used for anomaly detection.

TLDR: A method to embed content information within nodes in a GNN. Evaluated on imputation scenarios.

TLDR: Learn to adapt pasted snippets within some code context using transformers.

TLDR: Causal Discovery and Inference End-to-End.

TLDR: Generalize transformers and GNN to typed and qualified hypergraphs.

TLDR: A large Java dataset with compile-time metadata.

TLDR: Automatically generate (code) sketches, placing holes where ambiguity prevents us predicting terminal tokens.

TLDR: The natural language query is parsed and its structure instantiates neural modules that break down the search problem.

TLDR: Synthesize edit templates from edit sequences.

TLDR: Causal discovery and missing value imputation via GNNs.

TLDR: Learn seq2seq models that can edit sequence by copying long spans.

TLDR: Lightweight yet accurate code completion use neural reranking models.

TLDR: A survey of GNNs for learned program analyses

TLDR: Learn to detect a variety of bugs in source code by asking two models to play a hide-and-seek game: one model inserts a bug, the other tries to find it.

TLDR: Model code edits with tree-to-tree neural networks.

TLDR: Use meta-learning to predict Python type annotations, including rare ones.

TLDR: Automatically scraped code corpora commonly have many duplicates and evaluations are serverly affected by them.

TLDR: A benchmark for natural language code search using human-provided annotations.

TLDR: Generate code expressions using asynchronous GNNs.

TLDR: How can we represent edits in neural networks?

TLDR: Use code idioms to improve program synthesis and semantic parsing.

TLDR: A graph-to-sequence model for improved summarization of code and text.

TLDR: VAEs for graph encoding and generation.

TLDR: Sequence-based model to predict type annotations.

TLDR: Represent programs as graphs and use GNNs to find bugs.

TLDR: Automatically refine types, such as strings, respecting type constraints by using data flow and identifier names.

TLDR: Can we learn models that distiguish syntax from semantics of a math expression?

TLDR: Learn to adapt a snippet into new contexts.

TLDR: A 1D-CNN-to-sequence model to summarize source code.

TLDR: Coding conventions can be learned and suggested to developers.

TLDR: Mine interesting syntactic patterns in code.