跳到正文
arXiv cs.AI· Junqi Wang, Junyu Liu·· 4 小时前AI 评分18

用 QSP 构建可解的表征学习量子模型:精确核统计与可证明训练动力学

Representation Learning with Quantum Signal Processing

AI 导读

研究者将量子信号处理(QSP)确立为表征学习机制的可解量子模型,在任意深度下精确计算其量子神经正切核的均值与方差,发现对角项即使底层酉变换趋近 Haar 随机仍不自我平均。该工作进一步在无需冻结或集成平均核的情况下,为完整非线性梯度流证明稀疏数据保证,动力学收敛到带时变核闭包的标量流并给出显式收敛时间,同时存在有限深度速度上限。数值结果显示更高数据密度下出现超出标量流与冻结核描述的耦合演化。

正文 · 原文

View PDF HTML (experimental)

Abstract:Representation learning begins when training changes the features that define similarity between data. A frozen-kernel model only reweights a fixed geometry. We establish quantum signal processing (QSP) as a solvable quantum model of the representation-learning regime. At arbitrary depth, we compute the exact mean and variance of its quantum neural tangent kernel, revealing an input-dependent angular geometry whose diagonal remains non-self-averaging even when the underlying unitary approaches Haar randomness. We also prove a sparse-data guarantee for the full nonlinear gradient flow without freezing or ensemble-averaging the kernel: the realized dynamics converges to an integrable scalar flow with a time-dependent kernel closure and explicit convergence times. A finite-depth speed limit holds for every data set and trajectory. At higher data density, numerical results show coupled evolution beyond both the scalar and frozen-kernel descriptions. These results give a controlled theory of learned quantum data geometry with provable training dynamics beyond the frozen limit.
Comments: 7 pages, 2 figures (including End Matter); 16-page Supplemental Material with 1 figure. v2 improved presentation and expanded proof details in the Supplemental Material
Subjects: Quantum Physics (quant-ph); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2608.28828 [quant-ph]
  (or arXiv:2608.28828v2 [quant-ph] for this version)
  https://doi.org/10.48550/arXiv.2608.28828

arXiv-issued DOI via DataCite

Submission history

From: Junqi Wang [view email]
[v1] Fri, 28 Aug 2026 19:57:26 UTC (89 KB)
[v2] Mon, 16 Nov 2026 23:58:59 UTC (98 KB)

来源:arXiv cs.AI · arxiv.org