Backdoors in Neural Models of Source Code
Authors:
- 21 August 2022
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2892-2899
Abstract
Deep neural networks are vulnerable to a range of adversaries. A particularly pernicious class of vulnerabilities are backdoors, where model predictions diverge in the presence of subtle triggers in inputs. An attacker can implant a backdoor by poisoning the training data to yield a desired target prediction on triggered inputs. We study backdoors in the context of deep-learning for source code. (1) We define a range of backdoor classes for source-code tasks and install backdoors using dataset poisoning. (2) We adapt and improve recent algorithms from robust statistics for our setting, showing that backdoors leave a spectral signature in the learned representation of source code, thus enabling detection of poisoned data. (3) We conduct a thorough evaluation on different architectures and languages, showing the ease of injecting backdoors and our ability to eliminate them.Keywords
All Related Versions
Funding Information
- National Science Foundation
This publication has 19 references indexed in Scilit:
- code2vec: learning distributed representations of codeProceedings of the ACM on Programming Languages, 2019
- Deep learning type inference2018
- A Survey of Machine Learning for Big Code and NaturalnessACM Computing Surveys, 2018
- JBMC: A Bounded Model Checking Tool for Verifying Java Bytecode2018
- Satisfiability Modulo Theories2018
- Code completion with statistical language models2014
- Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation2014
- On the Relevance of Code Anomalies for Identifying Architecture Degradation Symptoms2012
- Anomaly detection using call stack information2004
- Tracking down software bugs using automatic anomaly detection2002