Research map / Architectures and efficiency / RNNs and normalization
Spectral Quantum Memory for Implicit Neural Representations
Daniel A Moroz-Broitman, Eliya Nachmani
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SQMM, a memory-coupled quantum implicit neural representation in which spectral components are produced sequentially by a shared quantum memory rather than independent quantum blocks.
Problem
Quantum implicit neural representations (QINRs) for signals treat spectral components as independent outputs of separate quantum blocks, leaving spectral capacity allocation poorly coupled.
Approach
A memory-coupled QINR (SQMM) in which sequential hybrid data re-uploading passes information across spectral components through a quantum memory, instead of computing each spectral component with an independent quantum block.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add shared quantum memory between bands (vs local qubits) (ablation) | SQMM · Cello | MSE (x1e-3): 0.53 -> 0.52 | No clear effect |
| add spectral memory alignment loss (ablation) | SQMM · Cello | MSE (x1e-3): 0.52 -> 0.48 | Improved |
| add spectral memory alignment loss (ablation) | SQMM · Astronaut | MSE (x1e-3): 1.23 -> 1.12 | Improved |
| add spectral memory alignment loss (ablation) | SQMM · Coffee | MSE (x1e-3): 0.88 -> 0.78 | Improved |
| add spectral memory alignment loss (ablation) | SQMM · Cello | LSD: 4.70 -> 4.45 | Improved |
| supervise final output spectrum (L_out) instead of memory spectrum (ablation) | SQMM · Cello | MSE (x1e-3): 0.54 vs 0.52 without loss | No clear effect |
| estimate expectations with 2048 shots instead of exact | SQMM · Cello | MSE (x1e-3): 0.48 -> 0.76 | Hurt |
| add more bands and memory qubits (ablation) | SQMM · Cello | MSE | Improved |
| increase gate angle noise | SQMM · Cello, images | MSE | Hurt |
Experimental setup
Builds on: quantum implicit neural representation data re-uploading
Closest papers by meaning
- Quantum Implicit Neural Representations (ICML 2024 · similarity 0.85)
- Sharp Capacity Scaling of Spectral Optimizers in Learning Associative Memory (NeurIPS 2026 · similarity 0.84)
- Quantum Visual Fields with Neural Amplitude Encoding (NeurIPS 2025 · similarity 0.84)
- Structured Unitary Tensor Network Representations for Circuit-Efficient Quantum Data Encoding (NeurIPS 2026 · similarity 0.84)
- Memory-Efficient Training with In-Place FFT Implementation (NeurIPS 2025 · similarity 0.82)
- A sampling theory perspective on activations for implicit neural representations (ICML 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
