thunlp/BkdAtk-LWS

Code and data of the ACL 2021 paper "Turn the Combination Lock: Learnable Textual Backdoor Attacks via Word Substitution"

38
/ 100
Emerging

This project helps researchers and developers working with natural language processing (NLP) models to understand and evaluate potential security vulnerabilities. It provides tools to create 'backdoor attacks' by subtly altering words in training data. The output is a "poisoned" NLP model that behaves normally until a specific, secret phrase is encountered, revealing the vulnerability. This is intended for NLP security researchers and AI safety engineers.

No commits in the last 6 months.

Use this if you need to research or demonstrate how textual backdoor attacks work on NLP models, or to test the robustness of your own models against such attacks.

Not ideal if you are looking for a general-purpose NLP library or a tool for data preprocessing or model training for typical tasks.

NLP-security AI-safety vulnerability-testing adversarial-machine-learning text-analysis
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 6 / 25
Maturity 16 / 25
Community 16 / 25

How are scores calculated?

Stars

16

Forks

6

Language

Python

License

MIT

Last pushed

Jun 29, 2021

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/nlp/thunlp/BkdAtk-LWS"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.