ExplAIner: A Declarative Query Language for Explaining Classification Models

2026-07-07 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Explainable AI · Depth: Expert, quick

Summary

ExplAIner is a new declarative query language designed for explaining predictions of Boolean classification models, building upon the existing FOIL language. Published on 2026-07-07, ExplAIner addresses two fundamental limitations of FOIL: its inability to express central optimality-based explanation queries and its high evaluation complexity over decision trees, which is hard for every level of the polynomial hierarchy. ExplAIner extends FOIL with an expanded vocabulary and layered structure, enabling it to express a broad range of explanation notions, including abductive, contrastive, feature-based, and distance-based queries. The evaluation problem for ExplAIner queries belongs to the Boolean hierarchy for Boolean models where basic predicates are polynomial-time evaluable, such as deterministic and decomposable Boolean circuits. Opt-FOIL, an optimization-oriented fragment for minimal explanations, has an evaluation problem in FP^NP. These findings indicate ExplAIner queries require a fixed number of SAT solver calls, and Opt-FOIL queries a polynomial number, practical for formal XAI.

Key takeaway

For AI Scientists developing or evaluating explainable AI systems, ExplAIner offers a robust framework to specify and compute diverse explanation notions. You should consider adopting declarative query languages like ExplAIner to unify your approach to abductive, contrastive, and feature-based explanations, especially when working with Boolean models. Its proven computational efficiency, requiring a fixed or polynomial number of SAT solver calls, means you can achieve practical, formal XAI without prohibitive overhead.

Key insights

ExplAIner offers a declarative language for diverse, computationally efficient ML model explanations.

Principles

• Declarative languages unify explanation notions.
• Optimality-based queries need specific language features.
• Fixed SAT solver calls enable practical XAI.

Method

ExplAIner extends FOIL with an extended vocabulary and layered structure to express various explanation notions. Opt-FOIL is an optimization-oriented fragment for minimal explanations.

In practice

• Use ExplAIner for diverse explanation types.
• Apply Opt-FOIL for minimal explanations.
• Leverage SAT solvers for efficient XAI.

Topics

• Explainable AI
• Declarative Query Languages
• Boolean Classification Models
• Computational Complexity
• SAT Solvers
• ExplAIner Language

Best for: Research Scientist, AI Scientist

Related on AIssential

• Bounded Fitting for Expressive Description Logics — Bounded fitting extends to expressive description logics while maintaining theoretical guarantees and practical performance.
• A Causal Argumentation Method for Explainability of Machine Learning Models — Integrating causal discovery with argumentation frameworks provides "why" explanations for ML model predictions.
• Explaining Neural Networks in Preference Learning: a Post-hoc Inductive Logic Programming Approach — ILASP can approximate black-box Neural Networks for preference learning, even in high-dimensional spaces.
• Explaining Multivariate Decision Trees: Characterising Tractable Languages — Finding abductive explanations for multivariate decision trees is tractable for specific feature languages, offering compact, explainable models.
• Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery — AI for mathematical reasoning is a rapidly evolving frontier, integrating diverse techniques for verified discovery.
• Interpretable and Explainable Surrogate Modeling for Simulations: A State-of-the-Art Survey and Perspectives on Explainable AI for Decision-Making — Integrating XAI with surrogate models enhances transparency and insight extraction from complex simulations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.