Structural Compositional Function Networks: Interpretable Functional Compositions for Tabular Discovery

TL;DR

StructuralCFN introduces an interpretable neural network architecture that models feature relationships explicitly through differentiable compositional functions, improving performance and interpretability on tabular data in scientific and clinical domains.

Contribution

The paper presents a novel architecture, StructuralCFN, that incorporates relation-aware inductive biases and adaptive gating to discover optimal feature compositions and inject domain knowledge.

Findings

Significant performance improvements on 18 benchmark datasets.

Provides human-readable symbolic expressions of data laws.

Achieves high interpretability with a compact model size.

Abstract

Despite the ubiquity of tabular data in high-stakes domains, traditional deep learning architectures often struggle to match the performance of gradient-boosted decision trees while maintaining scientific interpretability. Standard neural networks typically treat features as independent entities, failing to exploit the inherent manifold structural dependencies that define tabular distributions. We propose Structural Compositional Function Networks (StructuralCFN), a novel architecture that imposes a Relation-Aware Inductive Bias via a differentiable structural prior. StructuralCFN explicitly models each feature as a mathematical composition of its counterparts through Differentiable Adaptive Gating, which automatically discovers the optimal activation physics (e.g., attention-style filtering vs. inhibitory polarity) for each relationship. Our framework enables Structured Knowledge Integration, allowing domain-specific relational priors to be injected directly into the architecture to guide discovery. We evaluate StructuralCFN across a rigorous 10-fold cross-validation suite on 18 benchmarks, demonstrating statistically significant improvements (p < 0.05) on scientific and clinical datasets (e.g., Blood Transfusion, Ozone, WDBC). Furthermore, StructuralCFN provides Intrinsic Symbolic Interpretability: it recovers the governing "laws" of the data manifold as human-readable mathematical expressions while maintaining a compact parameter footprint (300--2,500 parameters) that is over an order of magnitude (10x--20x) smaller than standard deep baselines.