Alignment Unlocks Complementarity: A Framework for Multiview Circuit Representation Learning

2025-09-25

AI Summary

In multi-view learning for Boolean circuits, structural heterogeneity among graph representations—such as AIGs and XMGs—hinders effective fusion, while self-supervised masked modeling often misidentifies cross-view contextual information as noise. Method: We propose a “function-first alignment” paradigm, establishing functional equivalence as a prerequisite for multi-view self-supervised learning. We design an equivalence alignment loss to project heterogeneous graph structures into a shared, function-aware representation space; subsequently, we perform multi-view masked modeling with a staged curriculum training strategy. Contribution/Results: Our approach significantly enhances model generalization, transforming previously ineffective cross-view masked modeling into a strong pretraining signal. It achieves state-of-the-art (SOTA) performance on logic synthesis and combinational equivalence checking tasks, demonstrating robust transferability across downstream circuit understanding tasks.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Multi-modal Vision
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Multiview learning on Boolean circuits holds immense promise, as different graph-based representations offer complementary structural and semantic information. However, the vast structural heterogeneity between views, such as an And-Inverter Graph (AIG) versus an XOR-Majority Graph (XMG), poses a critical barrier to effective fusion, especially for self-supervised techniques like masked modeling. Naively applying such methods fails, as the cross-view context is perceived as noise. Our key insight is that functional alignment is a necessary precondition to unlock the power of multiview self-supervision. We introduce MixGate, a framework built on a principled training curriculum that first teaches the model a shared, function-aware representation space via an Equivalence Alignment Loss. Only then do we introduce a multiview masked modeling objective, which can now leverage the aligned views as a rich, complementary signal. Extensive experiments, including a crucial ablation study, demonstrate that our alignment-first strategy transforms masked modeling from an ineffective technique into a powerful performance driver.

Problem

• Overcoming structural heterogeneity between Boolean circuit representations
• Enabling effective multiview fusion for self-supervised learning
• Transforming cross-view context from noise to complementary signal

Innovation

• Functional alignment enables multiview self-supervision
• Equivalence Alignment Loss creates shared representation space
• Alignment-first curriculum unlocks masked modeling performance