Scalable Interconnect Learning in Boolean Networks

2025-07-03

AI Summary

Boolean networks suffer from explosive parameter growth in interconnection layers as input width increases, severely limiting scalability. Method: We propose a parameter-constant, differentiable, and trainable interconnection architecture—the first enabling end-to-end learning for wide-layer Boolean networks. Our approach jointly optimizes structural sparsity and functional equivalence by integrating SAT-solver–based logical equivalence pruning with activation-similarity–driven, data-aware pruning. Contribution/Results: Experiments demonstrate that our method achieves significant model compression—outperforming magnitude-based pruning baselines—while preserving high inference accuracy. It overcomes the fundamental width-scaling bottleneck of conventional Boolean networks, enabling scalable, verifiable, and compact neuro-symbolic models. This work establishes a new paradigm for efficient, formally tractable neural-symbolic integration.

Technology Category

• Machine Learning: Neuro-Symbolic Learning
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

Learned Differentiable Boolean Logic Networks (DBNs) already deliver efficient inference on resource-constrained hardware. We extend them with a trainable, differentiable interconnect whose parameter count remains constant as input width grows, allowing DBNs to scale to far wider layers than earlier learnable-interconnect designs while preserving their advantageous accuracy. To further reduce model size, we propose two complementary pruning stages: an SAT-based logic equivalence pass that removes redundant gates without affecting performance, and a similarity-based, data-driven pass that outperforms a magnitude-style greedy baseline and offers a superior compression-accuracy trade-off.

Problem

• Extend DBNs with scalable trainable interconnect
• Reduce model size via SAT-based logic pruning
• Improve compression-accuracy trade-off with data-driven pruning

Innovation

• Differentiable interconnect with constant parameters
• SAT-based pruning for redundant gate removal
• Data-driven similarity pruning for better compression