P30 - Möbius-Shapley: Native Feature Attribution for Quantum Logic
- Event
- iCCC2026 - iCampus Cottbus Conference
2026-05-05 - 2026-05-07
Cottbus - Band
- Poster
- Chapter
- Kommunikation
- Author(s)
- S. Dhahbi, I. Schmitt - BTU Cottbus-Senftenberg, Cottbus
- Pages
- 239 - 243
- DOI
- 10.5162/iCCC2026/P30
- ISBN
- 978-3-910600-10-2
- Price
- free
Abstract
The demand for transparent AI has made explainability crucial, particularly in high-stakes domains like healthcare and finance. Quantum Logic Decision Trees (QLDTs) offer interpretable classification but remain challenging for non-experts of logic to understand. Methods like SHAP provide model-agnostic explanations. However, there is a mismatch between the mintermbased QLDT’s logic of feature negation and SHAP’s assumption of feature missingness. We propose Möbius-Shapley, a native local (i.e., instance-specific) explanation method that bridges this gap by applying the Shapley value framework directly to Möbius-transformed QLDT weights. Our approach eliminates the semantic mismatch and provides native explanations, though we note that the exact derivation is computationally intensive for high-dimensional datasets compared to approximation methods.