Yin Zhenghao, Agresti Iris, de Felice Giovanni, Brown Douglas, Toumi Alexis, Pentangelo Ciro, Piacentini Simone, Crespi Andrea, Ceccarelli Francesco, Osellame Roberto, Coecke Bob, Walther Philip
University of Vienna, Faculty of Physics, Vienna Center for Quantum Science and Technology (VCQ), Vienna, Austria.
University of Vienna, Faculty of Physics, Vienna Doctoral School in Physics (VDSP), Vienna, Austria.
Nat Photonics. 2025;19(9):1020-1027. doi: 10.1038/s41566-025-01682-5. Epub 2025 Jun 2.
Recently, machine learning has had remarkable impact in scientific to everyday-life applications. However, complex tasks often require the consumption of unfeasible amounts of energy and computational power. Quantum computation may lower such requirements, although it is unclear whether enhancements are reachable with current technologies. Here we demonstrate a kernel method on a photonic integrated processor to perform a binary classification task. We show that our protocol outperforms state-of-the-art kernel methods such as gaussian and neural tangent kernels by exploiting quantum interference, and provides further improvements in accuracy by offering single-photon coherence. Our scheme does not require entangling gates and can modify the system dimension through additional modes and injected photons. This result gives access to more efficient algorithms and to formulating tasks where quantum effects improve standard methods.
最近，机器学习在从科学到日常生活的应用中都产生了显著影响。然而，复杂任务通常需要消耗大量难以实现的能量和计算能力。量子计算可能会降低此类要求，尽管目前尚不清楚当前技术是否能够实现性能提升。在此，我们展示了一种在光子集成处理器上执行二元分类任务的核方法。我们表明，我们的协议通过利用量子干涉优于诸如高斯核和神经切线核等当前最先进的核方法，并通过提供单光子相干性在精度上进一步提高。我们的方案不需要纠缠门，并且可以通过额外的模式和注入的光子来修改系统维度。这一结果为更高效的算法以及构建量子效应改进标准方法的任务开辟了道路。
