On the generalizability of Neural Program Models with respect to semantic-preserving program transformations (2020-07-31T00:00:00.000000Z)

TL;DR

The results show that even with small semantically preserving changes to the programs, these neural program models often fail to generalize their performance, and suggest that Neural program models based on data and control dependencies in programs generalize better than neural program model based only on abstract syntax trees (ASTs).

Authors

Mohammad Amin Alipour

7 papers

Lingxiao Jiang

3 papers

Md Rafiqul Islam Rabin

7 papers

Ke Wang

2 papers

Nghi D. Q. Bui

2 papers

Yijun Yu

1 papers

TL;DR

Authors

References73 items

Towards demystifying dimensions of source code embeddings

GNNGuard: Defending Graph Neural Networks against Adversarial Attacks

Blended, precise semantic program embeddings

Hoppity: Learning Graph Transformations to Detect and Fix Bugs in Programs

Evaluation of Generalizability of Neural Program Analyzers under Semantic-Preserving Transformations

Embedding Java Classes with code2vec: Improvements from Variable Obfuscation

Generating Adversarial Examples for Holding Robustness of Source Code Processing Models

Adversarial Robustness for Code

Semantic Robustness of Models of Source Code

Learning to Fix Build Errors with Graph2Diff Neural Networks

AutoFocus: Interpreting Attention-Based Neural Networks by Code Perturbation

Assessing the Generalizability of Code2vec Token Embeddings

Adversarial examples for models of code

Testing Neural Program Analyzers

Learning a Static Bug Finder from Data

Improving Attention Mechanism in Graph Neural Networks via Cardinality Preservation

Machine Learning Testing: Survey, Landscapes and Horizons

COSET: A Benchmark for Evaluating Neural Program Embeddings

Learning Scalable and Precise Representation of Program Semantics

Learning to Spot and Refactor Inconsistent Method Names

Software Engineering for Machine Learning: A Case Study

Maybe Deep Neural Networks are the Best Choice for Modeling Source Code

The adverse effects of code duplication in machine learning models of code

Modeling Attention Flow on Graphs

Deep learning type inference

Structured Neural Summarization

code2seq: Generating Sequences from Structured Representations of Code

Graph-to-Sequence Learning using Gated Graph Neural Networks

Adversarial Attack on Graph Structured Data

Generative Code Modeling with Graphs

Adversarial Attacks on Neural Networks for Graph Data

AI2: Safety and Robustness Certification of Neural Networks with Abstract Interpretation

Graph2Seq: Graph to Sequence Learning with Attention-based Neural Networks

code2vec: learning distributed representations of code

Adversarial Examples: Attacks and Defenses for Deep Learning

Dynamic Neural Program Embedding for Program Repair

Learning to Represent Programs with Graphs

Generating Natural Adversarial Examples

Graph Attention Networks

Boosting Adversarial Attacks with Momentum

A Survey of Machine Learning for Big Code and Naturalness

Are deep neural networks the best choice for modeling source code?

Adversarial Examples for Evaluating Reading Comprehension Systems

Axiomatic Attribution for Deep Networks

DeepFix: Fixing Common C Language Errors by Deep Learning

Understanding Neural Networks through Representation Erasure

Adversarial Machine Learning at Scale

Safety Verification of Deep Neural Networks

Towards Evaluating the Robustness of Neural Networks

sk_p: a neural program corrector for MOOCs

A Convolutional Attention Network for Extreme Summarization of Source Code

Gated Graph Sequence Neural Networks

DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks

Suggesting accurate method and class names

Recurrent Convolutional Neural Networks for Text Classification

Explaining and Harnessing Adversarial Examples

Convolutional Neural Networks over Tree Structures for Programming Language Processing

Neural Machine Translation by Jointly Learning to Align and Translate

Application of Deep Belief Networks for Natural Language Understanding

Structured Generative Models of Natural Source Code

Intriguing properties of neural networks

Multilayer feedforward networks are universal approximators

ModelingAttention Flow onGraphs. Relational Representation LearningWorkshop

Towards Verification of Artificial Neural Networks

Deep Learning

The Graph Neural Network Model

a wrong predicted label remains the same wrong label after the transformation

RQ3 How does the method length impact the generalizability of neural program models?

RQ1 How do the transformations impact the predictions of neural program models in the single-place transformed dataset?

RQ4 What are the trends in types of changes?

RQ2 When do the transformations affect neural program models

Based on the weighted average, GGNN performs worst for Switch to If and Unused Statement transformations. These two transformations add some additional nodes and paths in the AST

if models give less attention to those new paths, then the change can less effective

Field of Study

Journal Information

Name

Page