Kaiwu-PyTorch-Plugin: Bridging Deep Learning and Photonic Quantum Computing for Energy-Based Models and Active Sample Selection
Hongdong Zhu, Qi Gao, Yin Ma, Shaobo Chen, Haixu Liu, Fengao Wang, Tinglan Wang, Chang Wu, Kai Wen
TL;DR
The paper tackles the inefficiency of training energy-based models (EBMs) by leveraging photonic quantum computing via the Coherent Ising Machine (CIM) and integrating it into PyTorch. It introduces the Kaiwu-PyTorch-Plugin (KPP) with dual samplers (SA and CIM), an Active Sampling strategy formulated as a QUBO problem, and a model zoo including QBM, QBM-VAE, and Q-Diffusion to enable quantum-classical EBMs. Hardware-wise, it uses a CIM-based setup with a 1 km optical loop and cloud access to scale, while datasets span NLP, single-cell biology, and image-language tasks to demonstrate versatility. Empirically, the approach achieves SOTA results on single-cell representation benchmarks and improves perplexity on OpenWebText, validating a practical quantum-classical energy-based learning framework with potential for broader impact in scalable generative modeling and representation learning.
Abstract
This paper introduces the Kaiwu-PyTorch-Plugin (KPP) to bridge Deep Learning and Photonic Quantum Computing across multiple dimensions. KPP integrates the Coherent Ising Machine into the PyTorch ecosystem, addressing classical inefficiencies in Energy-Based Models. The framework facilitates quantum integration in three key aspects: accelerating Boltzmann sampling, optimizing training data via Active Sampling, and constructing hybrid architectures like QBM-VAE and Q-Diffusion. Empirical results on single-cell and OpenWebText datasets demonstrate KPPs ability to achieve SOTA performance, validating a comprehensive quantum-classical paradigm.