Understanding neural code intelligence through program simplification (2021-06-07T00:00:00.000000Z)

TL;DR

The proposed approach, Sivand, uses simplification techniques that reduce the size of input programs of a CI model while preserving the predictions of the model, and is broadly applicable across many model architectures and problem domains.

Abstract

A wide range of code intelligence (CI) tools, powered by deep neural networks, have been developed recently to improve programming productivity and perform program analysis. To reliably use such tools, developers often need to reason about the behavior of the underlying models and the factors that affect them. This is especially challenging for tools backed by deep neural networks. Various methods have tried to reduce this opacity in the vein of "transparent/interpretable-AI". However, these approaches are often specific to a particular set of network architectures, even requiring access to the network's parameters. This makes them difficult to use for the average programmer, which hinders the reliable adoption of neural CI systems. In this paper, we propose a simple, model-agnostic approach to identify critical input features for models in CI systems, by drawing on software debugging research, specifically delta debugging. Our approach, SIVAND, uses simplification techniques that reduce the size of input programs of a CI model while preserving the predictions of the model. We show that this approach yields remarkably small outputs and is broadly applicable across many model architectures and problem domains. We find that the models in our experiments often rely heavily on just a few syntactic features in input programs. We believe that SIVAND's extracted features may help understand neural CI systems' predictions and learned behavior.

Authors

Mohammad Amin Alipour

7 papers

Md Rafiqul Islam Rabin

7 papers

Vincent J. Hellendoorn

2 papers

TL;DR

Abstract

Authors

References45 items

Probing model signal-awareness via prediction-preserving input minimization

Towards demystifying dimensions of source code embeddings

On the generalizability of Neural Program Models with respect to semantic-preserving program transformations

Global Relational Models of Source Code

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

Toward Interpretable Machine Learning: Transparent Deep Neural Networks and Beyond

Adversarial Robustness for Code

Semantic Robustness of Models of Source Code

AutoFocus: Interpreting Attention-Based Neural Networks by Code Perturbation

Pre-trained Contextual Embedding of Source Code

Testing Neural Program Analyzers

Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailment

Generating Natural Language Adversarial Examples through Probability Weighted Word Saliency

How Often Do Single-Statement Bugs Occur? The ManySStuBs4J Dataset

Adversarial Examples Are Not Bugs, They Are Features

Neural Program Repair by Jointly Learning to Localize and Repair

Definitions, methods, and applications in interpretable machine learning

Deep learning type inference

code2seq: Generating Sequences from Structured Representations of Code

code2vec: learning distributed representations of code

Learning to Represent Programs with Graphs

A Survey of Machine Learning for Big Code and Naturalness

Attention is All you Need

Axiomatic Attribution for Deep Networks

The Shattered Gradients Problem: If resnets are the answer, then what is the question?

Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization

Evaluating non-adequate test-case reduction

Towards Evaluating the Robustness of Neural Networks

Deep API learning

A Convolutional Attention Network for Extreme Summarization of Source Code

Cause reduction: delta debugging, even without bugs

On the "Naturalness" of Buggy Code

Toward Deep Learning Software Repositories

Explaining and Harnessing Adversarial Examples

Cause Reduction for Quick Testing

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Visualizing and Understanding Convolutional Networks

On the naturalness of software

Vincent

Simplifying and Isolating Failure-Inducing Input

Multilayer feedforward networks are universal approximators

Demystifying Code Summarization Models

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics

Deep Learning

ESEC/FSE ’21, August 23–28, 2021, Athens, Greece

Field of Study

Journal Information

Name