Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reconciling Explanations in Multi-Model Systems through Probabilistic Argumentation

Shengxin Hong, Xiuyi Fan

TL;DR

Addresses the Explanation Reconciliation Problem (ERP) in multi-model AI systems by proposing a probabilistic-argumentation framework that maps models to probabilistic arguments and derives holistic explanations as a consistent distribution over related factors using a language $\mathcal{L}$ and rule set $\mathcal{R}$. Introduces the Relative Independence Assumption (RIA) to prune the search space and defines three user-oriented criteria—Optimistic, Pessimistic, and Laplace—for selecting explanations, with a reachability notion to identify related factors $\mathcal{L}_{\phi}$. Demonstrates computation of holistic explanations via constrained linear systems or maximum entropy optimization and discusses NP-hardness, providing a brain-disease diagnostic example to illustrate practical inferences. Positions the framework among prior work on multi-modal explanations and probabilistic logic, arguing that probabilistic argumentation enables coherent, model-wide explanations in high-stakes, interconnected AI systems.

Abstract

Explainable Artificial Intelligence (XAI) has become critical in enhancing the transparency and trustworthiness of AI systems, especially as these systems are increasingly deployed in high-stakes domains such as healthcare and finance. Despite the progress made in developing explanation generation techniques for individual machine learning (ML) models, significant challenges remain in achieving coherent and comprehensive explanations in multi-model systems. This paper addresses these challenges by focusing on the explanation reconciliation problem (ERP) within multi-model systems. Traditional explanation generation technique often fall short in multi-model systems contexts, where explanations from different models can conflict and fail to form a cohesive narrative. Through the use of probabilistic argumentation and knowledge representation techniques, we propose a framework for generating holistic explanations that align with human cognitive processes. Our approach involves mapping uncertain explanation information to probabilistic arguments and introducing criteria for explanation reconciliation based on user perspectives such as optimism, pessimism, fairness. In addition, we introduce the relative independence assumption to optimise the search space for computational explanations.

Reconciling Explanations in Multi-Model Systems through Probabilistic Argumentation

TL;DR

Addresses the Explanation Reconciliation Problem (ERP) in multi-model AI systems by proposing a probabilistic-argumentation framework that maps models to probabilistic arguments and derives holistic explanations as a consistent distribution over related factors using a language and rule set . Introduces the Relative Independence Assumption (RIA) to prune the search space and defines three user-oriented criteria—Optimistic, Pessimistic, and Laplace—for selecting explanations, with a reachability notion to identify related factors . Demonstrates computation of holistic explanations via constrained linear systems or maximum entropy optimization and discusses NP-hardness, providing a brain-disease diagnostic example to illustrate practical inferences. Positions the framework among prior work on multi-modal explanations and probabilistic logic, arguing that probabilistic argumentation enables coherent, model-wide explanations in high-stakes, interconnected AI systems.

Abstract

Explainable Artificial Intelligence (XAI) has become critical in enhancing the transparency and trustworthiness of AI systems, especially as these systems are increasingly deployed in high-stakes domains such as healthcare and finance. Despite the progress made in developing explanation generation techniques for individual machine learning (ML) models, significant challenges remain in achieving coherent and comprehensive explanations in multi-model systems. This paper addresses these challenges by focusing on the explanation reconciliation problem (ERP) within multi-model systems. Traditional explanation generation technique often fall short in multi-model systems contexts, where explanations from different models can conflict and fail to form a cohesive narrative. Through the use of probabilistic argumentation and knowledge representation techniques, we propose a framework for generating holistic explanations that align with human cognitive processes. Our approach involves mapping uncertain explanation information to probabilistic arguments and introducing criteria for explanation reconciliation based on user perspectives such as optimism, pessimism, fairness. In addition, we introduce the relative independence assumption to optimise the search space for computational explanations.
Paper Structure (6 sections, 3 theorems, 4 equations, 2 figures)

This paper contains 6 sections, 3 theorems, 4 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Background and Preliminaries
  3. Explanation Reconciliation
  4. Computing Holistic Explanation
  5. Related Works
  6. Conclusion

Key Result

Proposition 1

For any explanandum $\phi_i,\phi_k\in\mathcal{L}$ in $\mathcal{S}$, $\phi_i$ and $\phi_k$ are independent with each other.

Figures (2)

  • Figure 2: Illustration of Example 2. The nodes represent the model, the pointing arrows represent inputs and outputs of models.
  • Figure :

Theorems & Definitions (20)

  • Example 1
  • Definition 1
  • Definition 2
  • Definition 3
  • Definition 4
  • Definition 5
  • Example 2
  • Definition 6
  • Example 3
  • Definition 7
  • ...and 10 more