Reproduce “A Survey Of Methods For Explaining Black Box Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Riccardo Guidotti, Anna Monreale, Salvatore Ruggieri, Franco Turini, Dino Pedreschi, Fosca Giannotti

A survey by researchers at the University of Pisa and ISTI-CNR systematically reviews and classifies existing methods for explaining black box AI models, providing a taxonomy of problem types, explanation forms, and model characteristics. This work offers a structured understanding of the Explainable AI field and identifies key open research questions.

Key insights

  • The field of Explainable AI lacks a unified, formal definition for what constitutes an "explanation," hindering standardized evaluation and comparison of methods.
  • Most existing XAI methods are primarily designed for tabular data, with fewer solutions available for other complex data types such as sequences, graphs, or spatio-temporal data.
  • A
  • reverse engineering
  • paradigm, treating the black box as an oracle to train interpretable surrogate models, is a common methodological theme across many explanation problems.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.