A Foundation Model for Zero-Shot Logical Rule Induction
:::info Stub — Full Engineering Breakdown Coming This paper was featured on Hugging Face Daily Papers on 2026-05-06 with 3 upvotes. A full breakdown with production viability rating, implementation notes, and honest limitations is being written. Subscribe to AI Letters → :::
| Authors | Yin Jun Phua |
| Year | 2026 |
| HF Upvotes | 3 |
| arXiv | 2605.04916 |
| Download | |
| HF Page | View on Hugging Face |
Abstract
Inductive Logic Programming (ILP) learns interpretable logical rules from data. Existing methods are transductive: their learned parameters are bound to specific predicates and require retraining for each new task. We introduce Neural Rule Inducer (NRI), a pretrained model for zero-shot rule induction. Rather than encoding literal identities, NRI represents literals using domain-agnostic statistical properties such as class-conditional rates, entropy, and co-occurrence, which generalize across variable identities and counts without retraining. The model consists of a statistical encoder and a parallel slot-based decoder. Parallel decoding preserves the permutation invariance of logical disjunction; an autoregressive decoder would instead impose an arbitrary clause order. Product T-norm relaxation makes rule execution differentiable, allowing end-to-end training on prediction accuracy alone. We evaluate NRI on rule recovery, robustness to label noise and spurious correlations, and zero-shot transfer to real-world benchmarks, and we believe this work opens up the possibility of foundation models for symbolic reasoning. Code and the reference checkpoint are available at https://github.com/phuayj/neural-rule-inducer.
Engineering Breakdown
Plain English
This paper introduces Neural Rule Inducer (NRI), a pretrained foundation model that learns logical rules from data without retraining for new tasks. Instead of binding learned parameters to specific predicates, NRI encodes literals using domain-agnostic statistical properties like class-conditional rates and entropy, allowing the same model to generalize across different predicates and datasets zero-shot.
Key Engineering Insight
The core innovation is representing logical literals through statistical properties rather than literal identities, combined with parallel slot-based decoding instead of autoregressive generation. This preserves the mathematical structure of logical disjunction (permutation invariance) while enabling transfer across tasks without retraining.
Why It Matters for Engineers
For production ML systems, this eliminates the expensive retraining loop currently required when applying rule-learning to new domains or predicates. You can deploy one pretrained model across multiple data sources and use cases, significantly reducing engineering overhead and latency for rule discovery tasks like knowledge base construction or explainable decision systems.
Research Context
Traditional Inductive Logic Programming required task-specific parameter optimization, making it impractical for real-world deployment at scale. This work applies the foundation model paradigm—proven successful in NLP and vision—to symbolic reasoning, bridging the gap between interpretable rule learning and the transfer learning capabilities engineers expect from modern pretrained models.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::