Datasets:

Eciel
/

Smallest_k_experiment

license: mit
task_categories:
  - text-classification
language:
  - en
tags:
  - influence-functions
  - data-attribution
  - interpretability
pretty_name: Smallest-k Experiment Data

Smallest_k_experiment

Processed datasets and hyperparameter files for the paper "How Many and Which Training Points Would Need to be Removed to Flip this Prediction?" (Yang, Jain, Wallace; EACL 2023).

📄 Paper: https://aclanthology.org/2023.eacl-main.188/
💻 Code: https://github.com/ecielyang/Smallest_set

Summary

The paper finds a minimal subset of training points S_t whose removal would flip the prediction for a test point x_t, using two influence-function-based algorithms (IP and recursive_NT in the code repo). This dataset hosts the processed text-classification benchmarks (including BERT feature-extracted versions) and hyperparameter configs needed to reproduce those experiments.

Usage

git clone https://github.com/ecielyang/Smallest_set
# download data/hyperparameters from this repo, then:
mkdir results
python SST.py        # SST dataset
python SST_bert.py   # SST features from BERT

Files are serialized experiment artifacts, so the Dataset Viewer is disabled — download and load them directly per the code repo. English text classification; ~862 MB total.

Notes

Targets simple convex classifiers; results may not transfer to large non-convex models.
S_t is an approximation, not guaranteed globally minimal.

Citation

@inproceedings{yang-etal-2023-many,
    title = "How Many and Which Training Points Would Need to be Removed to Flip this Prediction?",
    author = "Yang, Jinghan and Jain, Sarthak and Wallace, Byron C.",
    booktitle = "Proceedings of the 17th Conference of the European Chapter of the ACL",
    year = "2023",
    url = "https://aclanthology.org/2023.eacl-main.188/",
    pages = "2571--2584",
}