Research Paper Co-authored by Associate Professor Shino Accepted at NeurIPS 2026

A research paper co-authored by Junnosuke Shino, Associate Professor of the School of International Liberal Studies, together with Kazuhiro Hiraki (International Monetary Fund), Shinichi Ishihara (Independent Researcher), and Takumi Kongo (Fukuoka University), titled “ESENSC: A Polynomial-Time Axiomatic Alternative to SHAP,” has been accepted to the Main Conference of NeurIPS 2026 (the Fortieth Annual Conference on Neural Information Processing Systems), a major international conference in machine learning and artificial intelligence.
NeurIPS is rated A*, the highest tier, in the ICORE 2026 conference rankings for computer science. It also ranks first in Artificial Intelligence and seventh overall in Google Scholar Metrics 2025, based on the h5-index.
SHAP, a method based on the Shapley value from cooperative game theory, is widely used to explain how much each input feature contributes to an AI model’s prediction. However, its computational cost increases rapidly as the number of features grows. Drawing on the axiomatic approach of cooperative game theory, this paper proposes a new attribution rule, ESENSC, that can be computed in polynomial time and shows that it is uniquely characterized by a set of desirable axioms. Experiments show that the proposed rule closely approximates SHAP while improving computational efficiency in high-dimensional settings.
NeurIPS 2026 will be held from December 6 to 12, 2026, in Sydney, Australia.

Paper (arXiv version): https://arxiv.org/abs/2603.00496
NeurIPS 2026: https://neurips.cc/Conferences/2026