Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neuro-Argumentative Learning with Case-Based Reasoning

Adam Gould, Francesca Toni

TL;DR

This paper addresses the challenge of interpretable reasoning in neural classifiers by proposing Gradual AA-CBR, a neuro-symbolic, end-to-end model that learns a data-point argumentation debate jointly with neural feature extractors. It replaces opaque NN reasoning with a differentiable edge-weighted QBAF whose gradual semantics yield argument strengths $\sigma(a)$ used to select class outcomes, enabling multi-class predictions and uncertainty quantification. Key contributions include automatic learning of feature and data-point importance, support for continuous data, and a transparent, debuggable reasoning process that matches NN performance while surpassing prior AA-CBR variants on binary tasks. The approach has practical impact for transparent decision-making in domains where human-aligned, case-based reasoning is valuable, with potential extensions to richer data modalities and explanations via explainable AI techniques.

Abstract

We introduce Gradual Abstract Argumentation for Case-Based Reasoning (Gradual AA-CBR), a data-driven, neurosymbolic classification model in which the outcome is determined by an argumentation debate structure that is learned simultaneously with neural-based feature extractors. Each argument in the debate is an observed case from the training data, favouring their labelling. Cases attack or support those with opposing or agreeing labellings, with the strength of each argument and relationship learned through gradient-based methods. This argumentation debate structure provides human-aligned reasoning, improving model interpretability compared to traditional neural networks (NNs). Unlike the existing purely symbolic variant, Abstract Argumentation for Case-Based Reasoning (AA-CBR), Gradual AA-CBR is capable of multi-class classification, automatic learning of feature and data point importance, assigning uncertainty values to outcomes, using all available data points, and does not require binary features. We show that Gradual AA-CBR performs comparably to NNs whilst significantly outperforming existing AA-CBR formulations.

Neuro-Argumentative Learning with Case-Based Reasoning

TL;DR

This paper addresses the challenge of interpretable reasoning in neural classifiers by proposing Gradual AA-CBR, a neuro-symbolic, end-to-end model that learns a data-point argumentation debate jointly with neural feature extractors. It replaces opaque NN reasoning with a differentiable edge-weighted QBAF whose gradual semantics yield argument strengths used to select class outcomes, enabling multi-class predictions and uncertainty quantification. Key contributions include automatic learning of feature and data-point importance, support for continuous data, and a transparent, debuggable reasoning process that matches NN performance while surpassing prior AA-CBR variants on binary tasks. The approach has practical impact for transparent decision-making in domains where human-aligned, case-based reasoning is valuable, with potential extensions to richer data modalities and explanations via explainable AI techniques.

Abstract

We introduce Gradual Abstract Argumentation for Case-Based Reasoning (Gradual AA-CBR), a data-driven, neurosymbolic classification model in which the outcome is determined by an argumentation debate structure that is learned simultaneously with neural-based feature extractors. Each argument in the debate is an observed case from the training data, favouring their labelling. Cases attack or support those with opposing or agreeing labellings, with the strength of each argument and relationship learned through gradient-based methods. This argumentation debate structure provides human-aligned reasoning, improving model interpretability compared to traditional neural networks (NNs). Unlike the existing purely symbolic variant, Abstract Argumentation for Case-Based Reasoning (AA-CBR), Gradual AA-CBR is capable of multi-class classification, automatic learning of feature and data point importance, assigning uncertainty values to outcomes, using all available data points, and does not require binary features. We show that Gradual AA-CBR performs comparably to NNs whilst significantly outperforming existing AA-CBR formulations.

Paper Structure

This paper contains 14 sections, 7 equations, 2 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background
  4. Methodology
  5. Gradual AA-CBR
  6. An End-to-End Neuro-Argumentative Learner
  7. Function Choices
  8. Model Training
  9. Experiments
  10. Experimental Results and Discussion
  11. Conclusion
  12. Gradual AA-CBR Training
  13. Experiment Details
  14. Visualising the Learned QBAF

Figures (2)

  • Figure 1: An example case-based debate generated by Gradual AA-CBR. Each node in the graph on the left-hand side is an argument represented by a data point in the training set. Red and green arrows indicate attacks and supports, respectively. The thickness of the argument borders and arrows represents the strength of the arguments and relationships. Argument importance and relationship strengths are computed with the feature extractor on the right-hand side.
  • Figure 2: A learned QBAF for the Iris dataset. Every argument in the casebase is a node in the graph, with the edges from each node representing attacks (in red) or supports (in green). We filter the edges to only those with a magnitude greater than 0.1 for visualisation purposes. The intensity of the colour indicates the strength of the attack or support.

Theorems & Definitions (4)

  • definition 1: Edge-Weighted Quantitative Bipolar Argumentation Framework
  • definition 2
  • definition 3: Gradual AA-CBR
  • definition 4: Regular Gradual AA-CBR