Answer Set Programming Energised! End-to-End Neurosymbolic Reasoning and Learning with ASP and Energy Based Models

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Expert, extended

Summary

ASPEn is a novel neurosymbolic reasoning and learning methodology that integrates Answer Set Programming (ASP) with Energy-Based Models (EBMs) for end-to-end training in dynamic domains. This approach supports joint optimization in a continuous latent space by fully incorporating ASP's declarative semantics, including background knowledge, constraints, and non-monotonic inference. It also provides a generalized platform for robust, ASP-centric training, particularly for applications involving perception and interaction. The methodology is practically implemented using Clingo for ASP solving and PyTorch for neural processing. Empirical evaluations demonstrate its effectiveness across various tasks: MNIST for digit-value assignment, achieving 98.67% and 98.81% per digit accuracy and 95.33% sum accuracy; CLEVR for compositional visual question-answering, yielding 62.85% VQA accuracy; and the MOT benchmark for multi-object tracking, showing comparable performance to similar trackers. ASPEn interprets ASP-based world models as structured embeddings, where stable models function as preferred configurations in a joint discrete-continuous space via semantically-guided energy minimization.

Key takeaway

For AI Scientists and Machine Learning Engineers developing robust systems for dynamic, real-world domains, ASPEn offers a principled neurosymbolic framework. You should consider this approach to integrate declarative knowledge representation with continuous optimization, especially for tasks requiring explainability and non-monotonic reasoning. This allows your models to jointly learn and reason, addressing perceptual uncertainty and spatio-temporal consistency. Explore its application for visual question-answering or multi-object tracking to enhance system trustworthiness and performance.

Key insights

ASPEn unifies Answer Set Programming and Energy-Based Models for robust, end-to-end neurosymbolic learning and reasoning in dynamic environments.

Principles

• ASP stable models define preferred configurations in an energy landscape.
• Logical constraints restrict feasible configurations, energy factors assign preferences.
• Non-monotonic reasoning generates structured learning signals.

Method

ASPEn grounds logic programs, evaluates neural energy functions for energised atoms, and performs MAP inference using Clingo's cost-based optimization to find minimal energy stable models. Learning uses contrastive divergence.

In practice

• Apply to visual question-answering (CLEVR).
• Use for multi-object tracking (MOT).
• Train digit classifiers with weak supervision (MNIST).

Topics

• Neurosymbolic AI
• Answer Set Programming
• Energy-Based Models
• Visual Question Answering
• Multi-Object Tracking
• Declarative Reasoning
• Machine Learning Integration

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Answer Set Programming Energised! End-to-End Neurosymbolic Reasoning and Learning with ASP and Energy Based Models — A neurosymbolic methodology integrates Answer Set Programming with Energy Based Models for robust, end-to-end reasoning and learning in dynamic domains.
• Answer Set Programming Energised! End-to-End Neurosymbolic Reasoning and Learning with ASP and Energy Based Models — End-to-end neurosymbolic reasoning integrates Answer Set Programming with Energy Based Models for robust learning in dynamic domains.
• A neurosymbolic Approach with Epistemic Deep Learning for Hierarchical Image Classification — A neurosymbolic framework enhances hierarchical image classification with epistemic uncertainty and logical consistency.
• Distilling Answer-Set Programming Rules from LLMs for Neurosymbolic Visual Question Answering — Distilling ASP rules from LLMs simplifies adapting logic-based VQA reasoning theories with minimal examples.
• 😺 🎙️ Watch: WTF is a "Reasoning Energy-Based Model"?! w/ Eve Bodnia of Logical Intelligence — Energy-based reasoning models offer a non-linguistic, efficient alternative to LLMs for complex problem-solving.
• Neurosymbolic Repo-level Code Localization — Neurosymbolic AI combining LLMs and Datalog overcomes keyword shortcuts for precise code localization.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

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