FIND: A Function Interpretation Dataset and Benchmark for Evaluating Interpretability Methods

Sarah Schwettmann, Tamar Rott Shaham, Joanna Materzyńska, Neil Chowdhury, Shuang� Li, Jacob Andreas, David Bau, Antonio TorralbaZenodo 2023 · DOI · OpenAlex

Abstract

<strong>FIND</strong> is an interactive dataset for evaluating AI interpretability methods on black box functions. This dataset contains all function files for the <strong>FIND</strong> benchmark and JSON files with associated metadata. The utilities provided in the associated <strong>FIND</strong> GitHub Repository support running and evaluating interpretation of the functions with user-defined interpreters.
BibTeX
@inproceedings{schwettmann2023find,
  author = {Sarah Schwettmann and Tamar Rott Shaham and Joanna Materzyńska and Neil Chowdhury and Shuang� Li and Jacob Andreas and David Bau and Antonio Torralba},
  title = {{FIND: A Function Interpretation Dataset and Benchmark for Evaluating Interpretability Methods}},
  booktitle = {Zenodo},
  year = {2023},
  doi = {10.5281/zenodo.8034162},
}

Loading related content...

Figure