Spectral Quantum Memory for Implicit Neural Representations

| Source: arXiv AI

Tags: quantum machine learning, implicit neural representations, SQMM, Fourier series, NeurIPS 2026

A NeurIPS 2026 paper introduces the Spectral Quantum Memory Model, a quantum implicit neural representation where layers share a coherent memory register, improving over compact baselines on audio and image reconstruction.

Details

Implicit neural representations map coordinates to signal values, and quantum versions usually treat spectral components as outputs of separate quantum blocks. SQMM links sequential hybrid data-reuploading bands through a shared coherent quantum memory register: each band reads it, computes on its coordinate upload and writes back part of its state. The authors prove the memory state forms an operator-valued Fourier series and that coupling creates no new primitive frequencies, which motivates a spectral memory alignment loss. On audio and image reconstruction benchmarks, SQMM improves over compact quantum and classical INR baselines. The paper is accepted to NeurIPS 2026.