KR2021Proceedings of the 18th International Conference on Principles of Knowledge Representation and ReasoningProceedings of the 18th International Conference on Principles of Knowledge Representation and Reasoning

Online event. November 3-12, 2021.

Edited by

ISSN: 2334-1033
ISBN: 978-1-956792-99-7

Sponsored by
Published by

Copyright © 2021 International Joint Conferences on Artificial Intelligence Organization

On the Identifiability of Hierarchical Decision Models

  1. Roman Bresson(Thales Research and Technology, Palaiseau, France, LISN CNRS-INRIA, Université Paris Saclay, France)
  2. Johanne Cohen(LISN CNRS-INRIA, Université Paris Saclay, France)
  3. Eyke Hüllermeier(University of Munich, Germany)
  4. Christophe Labreuche(Thales Research and Technology, Palaiseau, France, SINCLAIR, Industrial AI lab, Saclay, France)
  5. Michèle Sebag(LISN CNRS-INRIA, Université Paris Saclay, France)

Keywords

  1. Integrating knowledge representation and machine learning
  2. Decision making
  3. Knowledge-driven decision making
  4. Modeling and reasoning about preferences

Abstract

Interpretability is a desirable property for machine learning and decision models, particularly in the context of safety-critical applications. Another most desirable property of the sought model is to be unique or {\em identifiable} in the considered class of models: the fact that the same functional dependency can be represented by a number of syntactically different models adversely affects the model interpretability, and prevents the expert from easily checking their validity. This paper focuses on the Choquet integral (CI) models and their hierarchical extensions (HCI). HCIs aim to support expert decision making, by gradually aggregating preferences based on criteria; they are widely used

in multi-criteria decision aiding {and are receiving interest from the} Machine Learning {community}, as they preserve the high readability of CIs while efficiently scaling up w.r.t. the number of criteria.

The main contribution is to establish the identifiability property of HCI under mild conditions: two HCIs implementing the same aggregation function on the criteria space necessarily have the same hierarchical structure and aggregation parameters. The identifiability property holds even when the marginal utility functions are learned from the data. This makes the class of HCI models a most appropriate choice in domains where the model interpretability and reliability are of primary concern.