Universality of Many-body Projected Ensemble for Learning Quantum Data Distribution
Quoc Hoan Tran, Koki Chinzei, Yasuhiro Endo, and Hirotaka Oshima
TL;DR
This paper proves the universality of the Many-body Projected Ensemble (MPE) framework in approximating any quantum data distribution, providing theoretical guarantees and practical training methods validated by numerical experiments.
Contribution
It establishes a universality theorem for MPE in quantum state design and introduces an incremental training variant for improved practical learning.
Findings
MPE can approximate any pure state distribution within 1-Wasserstein distance.
Incremental MPE with layer-wise training enhances trainability.
Numerical results validate MPE's effectiveness on quantum datasets.
Abstract
Generating quantum data by learning the underlying quantum distribution poses challenges in both theoretical and practical scenarios, yet it is a critical task for understanding quantum systems. A fundamental question in quantum machine learning (QML) is the universality of approximation: whether a parameterized QML model can approximate any quantum distribution. We address this question by proving a universality theorem for the Many-body Projected Ensemble (MPE) framework, a method for quantum state design that uses a single many-body wave function to prepare random states. This demonstrates that MPE can approximate any distribution of pure states within a 1-Wasserstein distance error. This theorem provides a rigorous guarantee of universal expressivity, addressing key theoretical gaps in QML. For practicality, we propose an Incremental MPE variant with layer-wise training to improve…
Peer Reviews
Decision·Submitted to ICLR 2026
- Fundamental Theoretical Contribution: The primary strength of this paper is its main theorem. The question of universality is a cornerstone of any machine learning field, and providing a rigorous answer for quantum generative models is a significant achievement. This result provides a solid theoretical foundation for MPEs as a new class of quantum generative models. - Clear Empirical Validation: The experiments in Section 6 serve their purpose well. They are not intended to be state-of-the-ar
- Scope Limited to Pure States: The entire framework and theorem are limited to distributions over pure states. A fully general quantum generative model would also need to handle mixed states (density operators), which are the norm in noisy systems or when data comes from a subsystem. This is noted as a direction for future work but is a clear limitation of the current theorem. - Efficiency of the Universal Construction: The most significant limitation, which the authors commendably state in th
Many quantum machine learning papers are lack of theoretical guarantee, and due to the lack of good quantum hardware, it is hard to valuate the real performance. In this work, the author provide the proof for the universality of approximation, which provide some theoretical guarantee to the model.
However, the universality of approximation, is not really important for the quantum machine learning model, especially for the NISQ-friendly type. What matters is about the classical simulability, efficient training guarantee, size of parameters, quantum advantage... Especially many papers show negative results in recent years, I think these are more important to discuss, which is not covered by the authors. For example, the distribution generated by the many-body projected ensemble is purely cl
The paper is well written and has good experiments along with the theory.
While the paper has good results, I am not sure if universality result for a very specific QML architecture clears the bar for publication at a top ML conference like ICLR. May be a more specialized QML venue is more ideal.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsQuantum many-body systems · Quantum Computing Algorithms and Architecture · Machine Learning in Materials Science