Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An AI Architecture with the Capability to Explain Recognition Results

Paul Whitten, Francis Wolff, Chris Papachristou

TL;DR

The paper tackles the need for plain-language explanations in recognition systems by proposing an explainable architecture that combines explainable property-based flows with an unexplainable flow. It introduces two key contributions: the Ex_d explainability metric to quantify how explainable a decision is, and the E_PARS per-class effectiveness metric to improve accuracy, particularly in imbalanced datasets. Through experiments on MNIST and EMNIST, the authors show that integrating unexplainable yet high-performing flows boosts overall accuracy (often 2–10 percentage points, and up to around 98% with probability estimates) and that E_PARS consistently outperforms traditional metrics in per-class ranking. The work demonstrates the practical value of combining explainable components with selective unexplainable processes and highlights per-class metrics that better reflect model performance in real-world, imbalanced settings.

Abstract

Explainability is needed to establish confidence in machine learning results. Some explainable methods take a post hoc approach to explain the weights of machine learning models, others highlight areas of the input contributing to decisions. These methods do not adequately explain decisions, in plain terms. Explainable property-based systems have been shown to provide explanations in plain terms, however, they have not performed as well as leading unexplainable machine learning methods. This research focuses on the importance of metrics to explainability and contributes two methods yielding performance gains. The first method introduces a combination of explainable and unexplainable flows, proposing a metric to characterize explainability of a decision. The second method compares classic metrics for estimating the effectiveness of neural networks in the system, posing a new metric as the leading performer. Results from the new methods and examples from handwritten datasets are presented.

An AI Architecture with the Capability to Explain Recognition Results

TL;DR

The paper tackles the need for plain-language explanations in recognition systems by proposing an explainable architecture that combines explainable property-based flows with an unexplainable flow. It introduces two key contributions: the Ex_d explainability metric to quantify how explainable a decision is, and the E_PARS per-class effectiveness metric to improve accuracy, particularly in imbalanced datasets. Through experiments on MNIST and EMNIST, the authors show that integrating unexplainable yet high-performing flows boosts overall accuracy (often 2–10 percentage points, and up to around 98% with probability estimates) and that E_PARS consistently outperforms traditional metrics in per-class ranking. The work demonstrates the practical value of combining explainable components with selective unexplainable processes and highlights per-class metrics that better reflect model performance in real-world, imbalanced settings.

Abstract

Explainability is needed to establish confidence in machine learning results. Some explainable methods take a post hoc approach to explain the weights of machine learning models, others highlight areas of the input contributing to decisions. These methods do not adequately explain decisions, in plain terms. Explainable property-based systems have been shown to provide explanations in plain terms, however, they have not performed as well as leading unexplainable machine learning methods. This research focuses on the importance of metrics to explainability and contributes two methods yielding performance gains. The first method introduces a combination of explainable and unexplainable flows, proposing a metric to characterize explainability of a decision. The second method compares classic metrics for estimating the effectiveness of neural networks in the system, posing a new metric as the leading performer. Results from the new methods and examples from handwritten datasets are presented.
Paper Structure (11 sections, 9 equations, 2 figures, 10 tables)

This paper contains 11 sections, 9 equations, 2 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Explainable Architecture
  5. Explainable metric for Unexplainable Flows
  6. Effectiveness Metrics
  7. Inference Engine Training
  8. Results
  9. Explainable Results
  10. Effectiveness Results
  11. Explainable Example

Figures (2)

  • Figure 1: An explainable architecture with an unexplainable flow.
  • Figure 2: A letter 'S' example.