Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bound Propagation meets Constraint Simplification: Improving Logic-based XAI for Neural Networks

Ronaldo Gomes, Jairo Ribeiro, Luiz Queiroz, Thiago Alves Rocha

TL;DR

This work improves the efficiency of logic-based methods for explaining neural network decisions by combining bound propagation with constraint simplification, derived from the propagation, to tighten neuron bounds and eliminate unnecessary binary variables, making the explanation process more efficient.

Abstract

Logic-based methods for explaining neural network decisions offer formal guarantees of correctness and non-redundancy, but they often suffer from high computational costs, especially for large networks. In this work, we improve the efficiency of such methods by combining bound propagation with constraint simplification. These simplifications, derived from the propagation, tighten neuron bounds and eliminate unnecessary binary variables, making the explanation process more efficient. Our experiments suggest that combining these techniques reduces explanation time by up to 89.26\%, particularly for larger neural networks.

Bound Propagation meets Constraint Simplification: Improving Logic-based XAI for Neural Networks

TL;DR

This work improves the efficiency of logic-based methods for explaining neural network decisions by combining bound propagation with constraint simplification, derived from the propagation, to tighten neuron bounds and eliminate unnecessary binary variables, making the explanation process more efficient.

Abstract

Logic-based methods for explaining neural network decisions offer formal guarantees of correctness and non-redundancy, but they often suffer from high computational costs, especially for large networks. In this work, we improve the efficiency of such methods by combining bound propagation with constraint simplification. These simplifications, derived from the propagation, tighten neuron bounds and eliminate unnecessary binary variables, making the explanation process more efficient. Our experiments suggest that combining these techniques reduces explanation time by up to 89.26\%, particularly for larger neural networks.
Paper Structure (12 sections, 7 equations, 1 figure, 1 table, 1 algorithm)

This paper contains 12 sections, 7 equations, 1 figure, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. First-order Logic over LRA
  4. Classification Problems and Neural Networks
  5. MILP – Mixed Integer Linear Programming
  6. Logic-based XAI for NNs
  7. Our Approach
  8. Bound Propagation via Box for Logic-based XAI
  9. Constraint Simplification from Bound Propagation
  10. Experimental Results
  11. Results
  12. Conclusions

Figures (1)

  • Figure 1: Example of Neural Network

Theorems & Definitions (3)

  • Definition 1: Abductive Explanation
  • Example 1
  • Example 2