arXiv:2611.04574
Spectral Quantum Memory for Implicit Neural Representations
Daniel A. Moroz-Broitman, Eliya Nachmani
Abstract
Implicit neural representations (INRs) model signals as continuous functions from coordinates to values, but their performance depends strongly on how spectral capacity is allocated. Yet most Quantum INR (QINR) architectures treat spectral components as independent outputs of separate quantum blocks. We introduce the Spectral Quantum Memory Model (SQMM), a memory coupled quantum INR in which sequential hybrid data reuploading layers (bands) communicate through a shared coherent quantum memory register. Each band reads the memory, computes on its own coordinate upload, and writes part of its state back, so later bands can build on the computation of earlier ones. When memory read and write layers are input independent, we prove that the memory state forms an operator valued Fourier series whose support is contained in the Minkowski sum of the initial memory spectrum and the local band spectra. Downstream bands then condition their Fourier coefficients on the spectral content stored by earlier bands, and the memory coupling creates no new primitive frequencies. This theory motivates a spectral memory alignment loss that encourages memory to encode the spectrum expressed by the deployed prefix of the model. Across audio and image reconstruction benchmarks, SQMM improves over compact QINR and INR baselines.
AI Summary
The paper asks how quantum implicit neural representations can share spectral information across coordinate-processing blocks. Its Spectral Quantum Memory Model (SQMM) connects sequential data-reuploading bands through an unmeasured quantum memory: each band reads the memory, processes its coordinate upload, emits expectation-valued features, and updates the memory. A classical decoder combines the features. For input-independent memory couplings and finite Fourier expansions, a channel-expansion argument gives $\Omega_M^{(i)}\subseteq\Omega_M^{(0)}+\sum_{j=1}^{i}\Omega_j^{\mathrm{loc}}$, with analogous bounds for emitted features. Fourier coefficients propagate by convolution; coupling changes coefficients without adding primitive encoding frequencies. A training-only alignment loss matches normalized Fourier energies of memory Pauli probes to accumulated band-feature energies. Mixed-state simulations evaluate per-instance audio and image fitting, typically using five bands, two data qubits per band, two memory qubits, and roughly 650 parameters. Reported MSEs are lower than the evaluated compact quantum and classical baselines: Cello gives $0.48\times10^{-3}$ versus QIREN’s $6.79\times10^{-3}$, and grayscale CIFAR10 gives $0.46\times10^{-3}$ versus SIREN’s $0.98\times10^{-3}$. The parameter-matched independent-band Cello control gives $0.53\times10^{-3}$; memory without alignment gives $0.52\times10^{-3}$. The stated limitations include simulator cost, noise sensitivity, and measurement overhead; no hardware experiment or computational advantage is established. Sinusoidal positional encodings generally require lifted-coordinate analysis or finite spectral projections rather than exact finite raw-coordinate support. Alignment constrains only the sampled grid. Full memory probing scales as $\mathcal O(4^mT(n-1)/\epsilon^2)$ shots per training update. At 2048 shots, image errors remain above the strongest evaluated classical baselines.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- DOI
- 10.48550/arxiv.2611.04574
- Comments
- Accepted to Neurips 2026