The Standard Interpretable Model: A general theory of interpretable machine learning to deductively design interpretable methods using Lagrangian mechanics

TL;DR

A framework based on Lagrangian mechanics that deductively designs interpretable machine learning methods using a general theory.

Problem

Interpretability research suffers from fragmented methodologies and inconsistent evaluation protocols due to the lack of a general theory.

Approach

From user-defined interpretability premises, derive symmetries and constraints, then find optimal interpretable models via minima of a Lagrangian. Identify limitations of existing methods and propose new research directions.

Results & Contribution

Identifies and addresses limitations of existing methods (traditional, concept-based, mechanistic interpretability), contributes to interpretability education and research community.