ADVENT: LLM-Driven Automatic Predicate Invention for ILP

2026-07-02 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

ADVENT is an LLM-driven predicate invention (PI) mechanism designed to address a critical bottleneck in Inductive Logic Programming (ILP): the creation of new, semantically meaningful predicates. Traditional PI methods often require extensive domain expertise and generate opaque predicates, limiting their adaptability and reuse. ADVENT integrates Large Language Model abductive generation with Prolog deductive verification in an iterative loop, allowing concrete execution results to guide the LLM in refining candidate predicates. This mechanism identifies implicit patterns in structured relational data to invent auxiliary predicates with clear names and definitions. Invented predicates and learned rules are stored in a knowledge pool for cross-task reuse. Experiments on nine poker-hand concepts across seven LLMs demonstrated that LLM-driven PI achieved a 58% success rate where ILP alone failed, which increased to 80% with formal verification. The knowledge pool further yielded gains up to +31 percentage points, producing human-interpretable rules.

Key takeaway

For AI Scientists and NLP Engineers developing Inductive Logic Programming systems, ADVENT offers a path to overcome the predicate invention bottleneck. You should consider integrating LLM-driven abductive generation with formal deductive verification to create more interpretable and reusable predicates. This approach can significantly improve success rates in complex domains and build a valuable knowledge pool for future tasks, reducing reliance on manual domain expertise.

Key insights

ADVENT combines LLM abduction and Prolog deduction to automate predicate invention for ILP, enhancing interpretability and reuse.

Principles

• Iterative refinement improves LLM-generated predicates.
• Knowledge pools enable cross-task reuse.
• Formal verification boosts LLM output reliability.

Method

ADVENT employs an iterative loop where LLM abductive generation proposes predicates, and Prolog deductive verification refines them using concrete execution results, accumulating knowledge for reuse.

In practice

• Apply LLMs for pattern identification in relational data.
• Integrate deductive verification for AI-generated logic.
• Build knowledge pools for predicate reuse.

Topics

• Inductive Logic Programming
• Large Language Models
• Predicate Invention
• Abductive Reasoning
• Deductive Verification
• Knowledge Reuse

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Towards Automated Knowledge Transfer in Evolutionary Multitasking via Large Language Models — LLMs can autonomously design efficient and effective knowledge transfer models for multi-task optimization.
• FoVer: First-Order Logic Verification for Natural Language Reasoning — FoVer enhances LLM logical reasoning reliability via automated first-order logic verification using Z3.
• Structured Abductive-Deductive-Inductive Reasoning for LLMs via Algebraic Invariants — A symbolic reasoning scaffold improves LLM logical consistency by separating inference modes and enforcing algebraic invariants.
• As You Wish: Mission Planning with Formal Verification using LLMs in Precision Agriculture — Integrating LLMs with LTL formal verification enhances robotic mission planning in precision agriculture by resolving natural language ambiguity.
• Learning to Reason with Insight for Informal Theorem Proving — Teaching LLMs to identify core techniques significantly improves their informal theorem proving capabilities.
• AutoSpec: Safety Rule Evolution for LLM Agents via Inductive Logic Programming — AutoSpec evolves LLM agent safety rules using ILP-guided counterexample synthesis for interpretable, high-F1 performance.

Open in AIssential →

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