Search papers, labs, and topics across Lattice.
This paper introduces APEX, a model-attribution co-design framework that enables exact computation of Aumann-Shapley attributions in graph neural networks (GNNs) by leveraging a polynomial architecture. The key innovation is the PolyGIN architecture, which preserves a bounded multivariate polynomial form for model scores, allowing for efficient and exact evaluation of the attribution integral using Gauss鈥揕egendre quadrature. Experimental results demonstrate that APEX not only enhances attribution fidelity compared to existing methods but also reduces the computational burden associated with path integration in GNNs.
Exact Aumann-Shapley attributions in GNNs are now achievable with a polynomial architecture, drastically improving fidelity while slashing computational costs.
We study feature-level and node-level explanations for graph neural networks (GNNs) through the lens of Aumann-Shapley attribution. Path-integral methods such as Integrated Gradients provide an axiomatic formulation of attribution, but their practical use in deep GNNs typically relies on finite-sample numerical approximations to the path integral, requiring a trade-off between quadrature error and computational cost. This paper proposes APEX, a model-attribution co-design framework that makes the attribution integral exactly computable under a polynomial GNN architecture. The key component is PolyGIN, a GIN-style graph network whose message-passing, normalization, and transformation operations preserve a bounded multivariate polynomial form for scalar model scores, such as pre-softmax logits. We show that, for a PolyGIN with $L$ polynomial transformation blocks, the derivative along the attribution path has degree at most $2^L-1$. Therefore, Gauss--Legendre quadrature can evaluate the Aumann--Shapley path integral exactly, up to floating-point precision, with $2^{L-1}$ deterministic evaluation points. The resulting attributions can be computed at the feature level and then aggregated into node-level scores while preserving completeness. Experiments on synthetic and real-world graph benchmarks show that PolyGIN maintains competitive predictive performance, while the complete APEX framework achieves higher attribution fidelity than the compared baselines and substantially reduces the number of evaluations required for path integration.
