Frame Generation in Hilbert Space: Generative Interpolation of Measurement Data for Quantum Parameter Adaptation

Published in ICLR 2025 DeLTa Workshop on Deep Generative Models, 2025

This paper presents Generative Interpolation (GI), a novel method to reduce measurement overhead in Quantum Parameter Adaptation (QPA) for hybrid quantum-classical learning. Key contributions include:

  • Treating quantum measurement probabilities as analogous to video frames, enabling frame-wise estimation of unmeasured basis states.
  • A neural network-based generative model fills in missing measurement data, drastically reducing required quantum shots.
  • Enhanced fine-tuning performance for large language models (LLMs), demonstrating practical advantages for QPA under hardware constraints.

The proposed GI framework establishes a connection between deep generative learning and quantum measurement reconstruction, offering a scalable path forward for variational quantum algorithms.

Figure: Frame-wise generative interpolation model for quantum measurement completion in QPA tasks.

Recommended citation: Liu, Chen-Yu; Chen, Kuan-Cheng; Chen, Samuel Yen-Chi; Huang, Wei-Jia; & Chang, Yen Jui. (2025). "Frame Generation in Hilbert Space: Generative Interpolation of Measurement Data for Quantum Parameter Adaptation." ICLR 2025 Workshop on Deep Generative Models.
Download Paper