Explaining Neural Networks in Preference Learning: a Post-hoc Inductive Logic Programming Approach

2026-04-10 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A new approach proposes using Learning from Answer Sets (ILASP) to approximate black-box Neural Networks (NNs) in user preference learning. The method specifically explores ILASP's application to preference learning systems via weak constraints. Researchers created a dataset of user preferences for recipes to train NNs, which are then approximated by ILASP. Experiments evaluated ILASP as both a global and local approximator, addressing the challenge of high-dimensional feature spaces. To maintain transparency and manage computational time, a preprocessing step utilizing Principal Component Analysis (PCA) is introduced to reduce dataset dimensionality. This work is under consideration for publication in Theory and Practice of Logic Programming (TPLP).

Key takeaway

For Research Scientists developing explainable AI for preference learning, this work demonstrates a viable path to approximate complex Neural Networks using Inductive Logic Programming. You should consider integrating ILASP with a PCA preprocessing step to achieve transparent explanations without excessive computational overhead, especially when dealing with high-dimensional user preference data.

Key insights

ILASP can approximate black-box Neural Networks for preference learning, even in high-dimensional spaces.

Principles

• ILASP approximates NNs globally and locally.
• PCA reduces dimensionality while preserving transparency.

Method

The method involves training NNs on preference data, then using ILASP with weak constraints to approximate the NN's behavior. PCA is applied as a preprocessing step to reduce feature space dimensionality.

In practice

• Apply ILASP for NN explainability.
• Use PCA for high-dimensional preference data.

Topics

• Neural Networks
• Preference Learning
• Inductive Logic Programming
• ILASP
• Explainable AI

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning — ThoughtFold uses introspective preference learning to prune redundant steps in reasoning chains, significantly boosting LRM efficiency.
• IEEE Transactions on Evolutionary Computation, Volume 30, Issue 1, February 2025 — Evolutionary computation continues to advance, integrating with AI and deep learning for complex optimization and scheduling tasks.
• THEIA: Learning Complete Kleene Three-Valued Logic in a Pure-Neural Modular Architecture — Structured inductive bias is crucial for compositional generalization in neural networks learning three-valued logic.
• Honey, I Shrunk the Hypothesis Space (Through Logical Preprocessing) — Shrinking the hypothesis space pre-search dramatically reduces ILP learning times while maintaining accuracy.
• Explaining Multivariate Decision Trees: Characterising Tractable Languages — Finding abductive explanations for multivariate decision trees is tractable for specific feature languages, offering compact, explainable models.
• Representation Costs in Data Science: Foundations and the Quasi-Banach Spaces of Deep Neural Networks — A new framework unifies representation costs and native function spaces for parametric data-fitting methods, revealing deep neural network inductive biases.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.