TN-SHAP-G: Graph-Structured Tensor Network Surrogates for Shapley Values and Interactions

2026-06-01 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

TN-SHAP-G is a novel framework designed to efficiently compute Shapley values and higher-order interaction indices for black-box models with graph-structured inputs. Addressing the exponential computational cost of traditional Shapley value methods, TN-SHAP-G learns a compact, graph-aligned multilinear surrogate. This surrogate is represented as a tensor network, mirroring the input graph's topology, and is trained using a small number of oracle queries. Once trained, the framework enables deterministic recovery of first- and higher-order Shapley indices through the multilinear extension, eliminating the need for additional model queries or Monte Carlo variance. Published on 2026-06-01, experiments on molecular benchmarks demonstrate that TN-SHAP-G's learned factorization accurately matches exact Shapley values on small graphs and scales effectively to larger graphs where sampling-based approaches become impractical.

Key takeaway

For AI Scientists and Research Scientists working with black-box models on graph-structured data, TN-SHAP-G offers a significant advancement in interpretability. If you are struggling with the computational cost of traditional Shapley value methods for large graphs, this framework provides an efficient, deterministic alternative. You can achieve accurate first- and higher-order Shapley indices without extensive model queries or Monte Carlo variance, enabling scalable and reliable feature importance analysis.

Key insights

TN-SHAP-G efficiently computes Shapley values for graph-structured inputs using a graph-aligned tensor network surrogate.

Principles

• Exploit graph structure for Shapley computation.
• Tensor networks can mirror graph topology.
• Deterministic recovery avoids Monte Carlo variance.

Method

TN-SHAP-G learns a graph-aligned multilinear surrogate as a tensor network from a small number of oracle queries, then deterministically recovers Shapley indices via multilinear extension.

In practice

• Apply to molecular benchmarks for interpretability.
• Scale Shapley analysis to large graphs.

Topics

• Shapley Values
• Graph Neural Networks
• Tensor Networks
• Model Interpretability
• Black-box Models
• Feature Importance

Best for: AI Scientist, Research Scientist

Related on AIssential

• The Case for Model Science: Verify, Explore, Steer, Refine — The AI community needs a systematic "Model Science" to understand why complex models behave as they do, beyond benchmarking what they do.
• TriLens: Per-Layer Logit-Lens Entropy for White-Box Hallucination Detection — Internal model uncertainty, specifically per-layer logit-lens entropy, can effectively signal hallucination before final output.
• Beyond Task Success: Behavioral and Representational Diagnostics for WAM and VLA — WAMs improve object-level robot behavior, but architectural choices impact predictive representation and inference cost.
• Explainable deep reinforcement learning reveals energy-efficient control strategies for turbulent drag reduction — XDL-guided MARL significantly improves turbulent drag reduction and energy efficiency in fluid dynamics.
• Detection vs. Execution: Single-Bucket Probes Miss Half the Mamba-2 State Sink — Representational similarity in Mamba-2 does not imply functional equivalence for mechanistic interpretability probes.
• Tiny but Trusted: Efficient Vision-Language Reasoning for Time-Series Anomaly Detection — Fine-tuning VLMs with natural-language rationales significantly improves time-series anomaly detection performance and interpretability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.