Neuro-symbolic framework PACE generates feasibility-aware counterfactual explanations for ML models

A new paper proposes PACE, a neuro-symbolic framework designed to generate counterfactual explanations that are both plausible and actionable. The approach separates prediction and reasoning into two components: a neural predictive model for classification and a symbolic reasoning layer that enforces domain-specific constraints during counterfactual generation.

The framework is model-agnostic and adaptable to domains requiring realistic decision support. In a case study using the Adult Income dataset, the authors combined a multilayer perceptron classifier with Answer Set Programming (ASP) rules encoding feasible modifications to education, occupation, and working hours while preserving immutable attributes such as age and gender.

Results highlight a trade-off between counterfactual validity and plausibility, and show that symbolic constraints yield explanations that better satisfy domain-specific feasibility requirements. The authors argue this demonstrates the potential of neuro-symbolic methods for transparent, feasibility-aware counterfactual explanation in explainable AI.

The paper is available on arXiv as a first submission (v1) dated July 1, 2026, and includes four named authors: Pavel Iakovets, Liyanapathiranage Sudeepika Wajirakumari Samarathunga, Martin Thomas Horsch, and Fadi Al Machot.