Explanations Go Linear: Interpretable Meta-Encoding

The proposal Explanations Go Linear (Piaggesi et al., 2025) introduces an interpretable meta-encoding for post-hoc explanation of tabular models. It linearizes explanatory structure to produce stable, sparse, human-readable attribution patterns.

Unlike opaque embedding or surrogate stacking strategies, the meta-encoding preserves semantic alignment while improving reproducibility and speed—supporting rapid audit in time-sensitive contexts.

This contributes toward scalable inspection workflows for regulated decision pipelines.

References

2025

Explanations Go Linear: Post-hoc Explainability for Tabular Data with Interpretable Meta-Encoding

Simone Piaggesi, Riccardo Guidotti, Fosca Giannotti, and Dino Pedreschi

Dec 2025

RESEARCH LINE
1

Bib HTML
@misc{PGG2025, author = {Piaggesi, Simone and Guidotti, Riccardo and Giannotti, Fosca and Pedreschi, Dino}, line = {1}, month = dec, title = {Explanations Go Linear: Post-hoc Explainability for Tabular Data with Interpretable Meta-Encoding}, year = {2025} }