On the Decompositionality of Neural Networks

TL;DR

This paper introduces a formal notion of neural decompositionality based on semantic preservation of decision boundaries, and presents SAVED, a framework for boundary-aware neural decomposition.

Contribution

It defines neural decompositionality with a semantic criterion and develops SAVED, a framework to construct decompositions that preserve model decision boundary semantics.

Findings

Language Transformers largely preserve boundary semantics under decomposition.

Vision models frequently violate the decompositionality criterion.

Decompositionality is established as a formally definable and empirically testable property.

Abstract

Recent advances in deep neural networks have achieved state-of-the-art performance across vision and natural language processing tasks. In practice, however, most models are treated as monolithic black-box functions, limiting maintainability, component-wise optimization, and systematic testing and verification. Despite extensive work on pruning and empirical decomposition, the field still lacks a principled semantic notion of when a neural network can be meaningfully decomposed. We introduce neural decompositionality, a formal notion defined as a semantic-preserving abstraction over neural architectures. Our key insight is that decompositionality should be characterized by the preservation of semantic behavior along the model's decision boundary, which governs classification outcomes. This yields a semantic contract between the original model and its components, enabling a rigorous formulation of decomposition. Building on this foundation, we develop a boundary-aware framework, SAVED (Semantic-Aware Verification-Driven Decomposition), which operationalizes the proposed definition. SAVED combines counterexample mining over low logic-margin inputs, probabilistic coverage, and structure-aware pruning to construct decompositions that preserve decision-boundary semantics. We evaluate our approach on CNNs, language Transformers, and Vision Transformers. Results show clear architectural differences: language Transformers largely preserve boundary semantics under decomposition, whereas vision models frequently violate the decompositionality criterion, indicating intrinsic limits. Overall, our work establishes decompositionality as a formally definable and empirically testable property, providing a foundation for modular reasoning about neural networks.