Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BACON: A fully explainable AI model with graded logic for decision making problems

Haishi Bai, Jozo Dujmovic, Jianwu Wang

TL;DR

BACON presents a fully explainable AI framework that combines graded logic with a learnable LSP aggregation tree to enable end-to-end decision transparency. It replaces traditional neural aggregation with a GCD-based operator and uses a two-layer architecture (Gumbel-Sinkhorn permutation plus binary-tree aggregation) trained in two phases to discover interpretable feature interactions. The approach yields compact symbolic explanations, supports global and instance-level reasoning, and aligns with human decision-making, demonstrated on breast cancer diagnosis with 98.07% accuracy using 30 features and interpretable feature attribution and pruning. The work highlights practical implications for human-in-the-loop deployment, edge-device inference, and principled model refinement, while outlining future enhancements such as tree balancing, alternative aggregators, and cross-domain applications.

Abstract

As machine learning models and autonomous agents are increasingly deployed in high-stakes, real-world domains such as healthcare, security, finance, and robotics, the need for transparent and trustworthy explanations has become critical. To ensure end-to-end transparency of AI decisions, we need models that are not only accurate but also fully explainable and human-tunable. We introduce BACON, a novel framework for automatically training explainable AI models for decision making problems using graded logic. BACON achieves high predictive accuracy while offering full structural transparency and precise, logic-based symbolic explanations, enabling effective human-AI collaboration and expert-guided refinement. We evaluate BACON with a diverse set of scenarios: classic Boolean approximation, Iris flower classification, house purchasing decisions and breast cancer diagnosis. In each case, BACON provides high-performance models while producing compact, human-verifiable decision logic. These results demonstrate BACON's potential as a practical and principled approach for delivering crisp, trustworthy explainable AI.

BACON: A fully explainable AI model with graded logic for decision making problems

TL;DR

BACON presents a fully explainable AI framework that combines graded logic with a learnable LSP aggregation tree to enable end-to-end decision transparency. It replaces traditional neural aggregation with a GCD-based operator and uses a two-layer architecture (Gumbel-Sinkhorn permutation plus binary-tree aggregation) trained in two phases to discover interpretable feature interactions. The approach yields compact symbolic explanations, supports global and instance-level reasoning, and aligns with human decision-making, demonstrated on breast cancer diagnosis with 98.07% accuracy using 30 features and interpretable feature attribution and pruning. The work highlights practical implications for human-in-the-loop deployment, edge-device inference, and principled model refinement, while outlining future enhancements such as tree balancing, alternative aggregators, and cross-domain applications.

Abstract

As machine learning models and autonomous agents are increasingly deployed in high-stakes, real-world domains such as healthcare, security, finance, and robotics, the need for transparent and trustworthy explanations has become critical. To ensure end-to-end transparency of AI decisions, we need models that are not only accurate but also fully explainable and human-tunable. We introduce BACON, a novel framework for automatically training explainable AI models for decision making problems using graded logic. BACON achieves high predictive accuracy while offering full structural transparency and precise, logic-based symbolic explanations, enabling effective human-AI collaboration and expert-guided refinement. We evaluate BACON with a diverse set of scenarios: classic Boolean approximation, Iris flower classification, house purchasing decisions and breast cancer diagnosis. In each case, BACON provides high-performance models while producing compact, human-verifiable decision logic. These results demonstrate BACON's potential as a practical and principled approach for delivering crisp, trustworthy explainable AI.

Paper Structure

This paper contains 33 sections, 4 equations, 11 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. End-to-end explainability
  3. Human-AI collaboration
  4. Background and Related Works
  5. Automated symbolic intelligence for explainable AI
  6. Graded Logic and the LSP decision method
  7. BACON architecture
  8. BACON model architecture
  9. Training process
  10. How BACON delivers end-to-end explainability
  11. Feature attribution
  12. Compositional transparency and full decision-making pathway
  13. Full alignment with human decision-making process
  14. BACON in breast cancer diagnosis use case
  15. Breast cancer diagnosis
  16. ...and 18 more sections

Figures (11)

  • Figure 1: (1) A graded propositional logic model structured as a binary tree; (2) A sample job selection LSP tree
  • Figure 2: High-level architecture of BACON
  • Figure 3: Feature attribution with BACON: (1) important vs. irrelevant features; (2) conjunction; (3) disjunction; (4) near drastic conjunction; (5) near drastic disjunction; (6) neutral; (7) weighted disjunction
  • Figure 4: Breast cancer diagnosis process modeled as LSP aggregation tree
  • Figure 5: Converting to degree of truth to binary output (threshold = 0.5)
  • ...and 6 more figures