TM4QEM: Low-Regret Clause-Based Tsetlin Policies for Cost-Aware Quantum Error-Mitigation Selection
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 725-732
DOI: 10.1609/aaaiss.v10i1.43481 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Krishna Bhatia und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Quantum error-mitigation (QEM) techniques offer different error reduction and execution overhead across circuit and noise regimes, so choosing a mitigation method is itself a decision problem. We formulate QEM orchestration as finite-action utility learning and introduce TM4QEM, a Tsetlin-based selector in which native propositional clause activations are used to predict the utility of every mitigation action. This differs from direct multiclass Tsetlin prediction: near-tied actions are compared by estimated utility, and a cost term is applied only at inference, allowing the same learned model to serve multiple overhead preferences. We evaluate eleven controlled QEM-action proxies on five independently generated 3,000-instance datasets, using both stratified random splits and a held-out highest-noise quartile. At zero cost weight, direct weighted-TM action classification obtains mean regret 0.00417 on random splits and 0.00984 under high-noise shift. Clause-utility modelling reduces these to 0.00266 and 0.00614, reductions of 36.2% and 37.6%. A regret-aware clause bank reaches 0.00253 on random splits but does not transfer, degrading to 0.00710 under high noise; the simpler clause-utility policy is the better default. XGBoost remains the strongest overall reference (0.00148/0.00521), so we do not claim predictive dominance. Instead, the results identify a more defensible role for Tsetlin policies: interpretable action-utility decomposition with low regret, explicit cost control, and inspectable failure modes. Sparse and margin-aware extensions do not consistently improve the simpler clause-utility policy, providing useful negative evidence about the accuracy–compactness trade-off. The study is simulator-derived and evaluates policy selection, not physical superiority of the named QEM methods.
Zitieren
APA
Bhatia, K., & Bajaj, S. (2026). TM4QEM: Low-Regret Clause-Based Tsetlin Policies for Cost-Aware Quantum Error-Mitigation Selection. Proceedings of the AAAI Symposium Series, 10(1), 725-732. https://doi.org/10.1609/aaaiss.v10i1.43481
Vancouver
Bhatia K, Bajaj S. TM4QEM: Low-Regret Clause-Based Tsetlin Policies for Cost-Aware Quantum Error-Mitigation Selection. Proceedings of the AAAI Symposium Series. 2026;10(1):725-732. doi:10.1609/aaaiss.v10i1.43481
BibTeX
@article{Bhatia2026,
author = {Bhatia, Krishna and Bajaj, Smriti},
title = {TM4QEM: Low-Regret Clause-Based Tsetlin Policies for Cost-Aware Quantum Error-Mitigation Selection},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {725--732},
doi = {10.1609/aaaiss.v10i1.43481}
}
EUVIMED