nlx-group/Shortcutted-Commonsense-Reasoning

Code for the article "Shortcutted Commonsense: Data Spuriousness in Deep Learning of Commonsense Reasoning", Outstanding Paper at EMNLP2021

21
/ 100
Experimental

This project helps AI researchers and practitioners evaluate if their deep learning models for commonsense reasoning are truly learning robust generalizations or just exploiting superficial patterns in their training data. You provide your Transformer models and commonsense datasets, and the project outputs insights into whether your models are truly understanding the underlying problems or relying on "shortcuts." It's for anyone building or evaluating AI systems that require human-like common sense.

No commits in the last 6 months.

Use this if you are a researcher or AI engineer who wants to rigorously test the robustness and genuine reasoning capabilities of your natural language processing models, especially those built on Transformers for commonsense tasks.

Not ideal if you are looking for an out-of-the-box solution to directly improve model performance without an interest in deep analytical insights into data spuriousness or shortcut learning.

AI research natural language understanding model interpretability dataset analysis machine learning evaluation
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 5 / 25
Maturity 16 / 25
Community 0 / 25

How are scores calculated?

Stars

10

Forks

Language

Jupyter Notebook

License

MIT

Last pushed

Nov 07, 2021

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/nlx-group/Shortcutted-Commonsense-Reasoning"

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