princeton-nlp/rationale-robustness
NAACL 2022: Can Rationalization Improve Robustness? https://arxiv.org/abs/2204.11790
This project helps researchers and practitioners evaluate and improve the reliability of natural language processing (NLP) models, especially when these models need to explain their decisions. It takes text data and an existing NLP model's predictions, then tests how well the model's explanations (rationales) prevent it from being tricked by irrelevant or 'attack' text. The output shows how robust the model's explanations are in different scenarios, which is useful for NLP researchers and AI ethics practitioners.
No commits in the last 6 months.
Use this if you are working with NLP models that provide explanations for their predictions and need to understand how resilient these explanations are to adversarial inputs or 'noisy' text.
Not ideal if you are looking for a general-purpose NLP model or if your primary concern is model accuracy rather than the robustness of its explanations.
Stars
27
Forks
1
Language
Python
License
—
Category
Last pushed
Nov 21, 2022
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/nlp/princeton-nlp/rationale-robustness"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
thunlp/OpenAttack
An Open-Source Package for Textual Adversarial Attack.
thunlp/TAADpapers
Must-read Papers on Textual Adversarial Attack and Defense
jind11/TextFooler
A Model for Natural Language Attack on Text Classification and Inference
thunlp/OpenBackdoor
An open-source toolkit for textual backdoor attack and defense (NeurIPS 2022 D&B, Spotlight)
thunlp/HiddenKiller
Code and data of the ACL-IJCNLP 2021 paper "Hidden Killer: Invisible Textual Backdoor Attacks...